Stampede2 User Guide
Last update: September 21, 2017 see revision history

Notices

  • Stampede2 has its own home and scratch file systems and a new $WORK directory. You'll need to transfer your files from Stampede1 to Stampede2. See Managing Files for information that will help you do so easily.
  • Stampede2's software stack is newer than the software on the decommissioned Stampede1 KNL sub-system. Be sure to recompile before running on Stampede2. See Building Software for more information.
  • Stampede2's accounting system is based on node-hours: one Service Unit (SU) represents a single compute node used for one hour (a node-hour) rather than a core-hour. See Job Accounting for more information.
  • Stampede2's KNL nodes have 68 cores, each with 4 hardware threads. But it may not be a good idea to use all 272 hardware threads simultaneously, and it's certainly not the first thing you should try. In most cases it's best to specify no more than 64-68 MPI tasks or independent processes per node, and 1-2 threads/core. See Best Known Practices for more information.

Figure 1. Stampede2 System

Introduction

Stampede2, generously funded by the National Science Foundation (NSF) through award ACI-1134872, is the flagship supercomputer at the Texas Advanced Computing Center (TACC), University of Texas at Austin. It will enter full production in the Fall 2017 as an 18-petaflop national resource that builds on the successes of the original Stampede system it replaces. The first phase of the Stampede2 rollout features the second generation of processors based on Intel's Many Integrated Core (MIC) architecture. Stampede2's 4,200 Knights Landing (KNL) nodes represent a radical break with the first-generation Knights Corner (KNC) MIC coprocessor. Unlike the legacy KNC, a Stampede2 KNL is not a coprocessor: each 68-core KNL is a stand-alone, self-booting processor that is the sole processor in its node. The Phase 1 KNL system is now available. Later this summer Phase 2 will add to Stampede2 a total of 1,736 Intel Xeon Skylake (SKX) nodes.

The older Stampede system (sometimes "Stampede1" for simplicity) will remain in production until Fall 2017. We are gradually reducing its size to make room for Stampede2 components. We will not decommission Stampede1, however, until Stampede2 is in full production; the two systems will coexist for an extended transition period.

Phase 1 Compute Nodes (KNL)

Stampede2 hosts 4,200 KNL compute nodes, including 504 KNL nodes that were formerly configured as a Stampede1 sub-system.

Table 1. Stampede2 KNL Compute Node Specifications

Model:  Intel Xeon Phi 7250 ("Knights Landing")
Total cores per KNL node:  68 cores on a single socket
Hardware threads per core:  4
Hardware threads per node:  68 x 4 = 272
Clock rate:  1.4GHz
RAM:  96GB DDR4 plus 16GB high-speed MCDRAM. Configurable in two important ways; see Programming and Performance for more info.
Cache:  32KB L1 data cache per core; 1MB L2 per two-core tile. In default config, MCDRAM operates as 16GB direct-mapped L3.
Local storage:  All but 504 KNL nodes have a 132GB /tmp partition on a 200GB Solid State Drive (SSD). The 504 KNLs originally installed as the Stampede1 KNL sub-system each have a 58GB /tmp partition on 112GB SSDs. The latter nodes currently make up the development, flat-quadrant and flat-snc4 queues.

Each of Stampede2's KNL nodes includes 96GB of traditional DDR4 Random Access Memory (RAM). They also feature an additional16GB of high bandwidth, on-package memory known as Multi-Channel Dynamic Random Access Memory (MCDRAM) that is up to four times faster than DDR4. The KNL's memory is configurable in two important ways: there are BIOS settings that determine at boot time the processor's memory mode and cluster mode. The processor's memory mode determines whether the fast MCDRAM operates as RAM, as direct-mapped L3 cache, or as a mixture of the two. The cluster mode determines the mechanisms for achieving cache coherency, which in turn determines latency: roughly speaking, this mode specifies the degree to which some memory addresses are "closer" to some cores than to others. See Programming and Performance below for a top-level description of these and other available memory and cluster modes.

Phase 2 Compute Nodes (SKX)

When Phase 2 is complete, Stampede2 will host 1,736 SKX nodes.

Table 2. Stampede2 SKX Compute Node Specifications

Model:  Intel Xeon Platinum 8160 ("Skylake")
Total cores per SKX node:  48 cores on two sockets (24 cores/socket)
Hardware threads per core:  2
Hardware threads per node:  48 x 2 = 96
Clock rate:  2.1GHz
RAM:  192GB (2.67GHz)
Cache:  32KB L1 data cache per core; 1MB L2 per core; 33MB L3 per socket. Each socket can cache up to 57MB (sum of L2 and L3 capacity).
Local storage:  132 /tmp partition on a 200GB SSD

Network

The interconnect is a 100Gb/sec Intel Omni-Path (OPA) network with a fat tree topology employing six core switches. There is one leaf switch for each 28-node half rack, each with 20 leaf-to-core uplinks (28/20 oversubscription).

File Systems Introduction

Stampede2 mounts three shared Lustre file systems on which each user has corresponding account-specific directories $HOME, $WORK, and $SCRATCH. Each file system is available from all Stampede2 nodes; the Stockyard-hosted work file system is available on other TACC systems as well. A Lustre file system looks and acts like a single logical hard disk, but is actually a sophisticated integrated system involving many physical drives (dozens of physical drives for $HOME, hundreds for $SCRATCH, and thousands for $WORK).

Lustre can stripe (distribute) large files over several physical disks, making it possible to deliver the high performance needed to service input/output (I/O) requests from hundreds of users across thousands of nodes. Object Storage Targets (OSTs) manage the file system's spinning disks: a file with 20 stripes, for example, is distributed across 20 OSTs. One designated Meta-Data Server (MDS) tracks the OSTs assigned to a file, as well as the file's descriptive data.

See Navigating the Shared File Systems and consult Shared Lustre File Systems in the Citizenship sections for best practices.

Table 3. Stampede2 File Systems

File System Quota Key Features
$HOME 10GB, 200,000 files Not intended for parallel or high-intensity file operations.
Backed up regularly.
Overall capacity ~1PB. Two Meta-Data Servers (MDS), four Object Storage Targets (OSTs). Defaults: 1 stripe, 1MB stripe size.
$WORK 1TB, 3,000,000 files across all TACC systems,
regardless of where on the file system the files reside.
Not intended for high-intensity file operations or jobs involving very large files.
On the Global Shared File System that is mounted on most TACC systems.
See Stockyard system description for more information.
Not backed up.
$SCRATCH no quota Overall capacity ~30PB. Four MDSs, 66 OSTs. Defaults: 1 stripe, 1MB stripe size.
Not backed up.

Specialized Nodes

We plan to add large-memory nodes in 2018. There are no plans to add Graphics Processing Units (GPUs) to the system.

Accessing the System

Access to all TACC systems now requires Multi-Factor Authentication (MFA). You can create an MFA pairing on the TACC User Portal. After login on the portal, go to your account profile (Home->Account Profile), then click the "Manage" button under "Multi-Factor Authentication" on the right side of the page. See Multi-Factor Authentication at TACC for further information.

Secure Shell (SSH)

The "ssh" command (SSH protocol) is the standard way to connect to Stampede2. SSH also includes support for the file transfer utilities scp and sftp. Wikipedia is a good source of information on SSH. SSH is available within Linux and from the terminal app in the Mac OS. If you are using Windows, you will need an SSH client that supports the SSH-2 protocol: e.g. Bitvise, OpenSSH, PuTTY, or SecureCRT. Initiate a session using the ssh command or the equivalent; from the Linux command line the launch command looks like this:

localhost$ ssh myusername@stampede2.tacc.utexas.edu

The above command will rotate connections across all available login nodes and route your connection to one of them. To connect to a specific login node, use its full domain name:

localhost$ ssh myusername@login2.stampede2.tacc.utexas.edu

To connect with X11 support on Stampede2 (usually required for applications with graphical user interfaces), use the "-X" or "-Y" switch:

localhost$ ssh -X myusername@stampede2.tacc.utexas.edu

Use your TACC password, not your XSEDE password, for direct logins to TACC resources. You can change your TACC password through the TACC Portal. Select "Change Password" under the "HOME" tab after login. If you've forgotten your password, go to the TACC User Portal home page and select the "Forgot your password?" link in the login area.

To report a connection problem, execute the ssh command with the "-vvv" option and include the verbose output when submitting a help ticket.

Do not run the "ssh-keygen" command on Stampede2. This command will create and configure a key pair that will interfere with the execution of job scripts in the batch system. If you do this by mistake, you can recover by renaming or deleting the .ssh directory located in your home directory. One good way to do this:

  1. execute "mv .ssh dot.ssh.old"
  2. log out
  3. log back in

After logging in again the system will generate a properly configured key pair.

GSI-OpenSSH (gsissh)

GSI-OpenSSH is a customized implementation of OpenSSH that includes support for Grid Security Infrastructure (GSI) authentication and credential forwarding.

Use port 2222 for gsissh connections to Stampede2. The following sequence of commands authenticate using the XSEDE myproxy server, then connects to Stampede2 via this port:

localhost$ myproxy-logon -s myproxy.teragrid.org
localhost$ gsissh -p 2222 userid@stampede2.tacc.utexas.edu

XSEDE Single Sign-On Hub

XSEDE users can also access Stampede2 via the XSEDE Single Sign-On Hub.

When reporting a problem to the help desk, please execute the gsissh command with the "-vvv" option and include the verbose output in your problem description.

Using Stampede2

Stampede2 nodes run Linux (Red Hat Enterprise Linux 7). Regardless of your research workflow, you will almost certainly need to be comfortable with Linux basics and a Linux-based text editor (e.g. vi, emacs, nano, or gedit). There are numerous resources in a variety of formats that available to help you acquire this understanding, including some listed on the TACC and XSEDE training sites. This user guide presumes you have that understanding. If you encounter a term or concept in the material below that is new to you, a quick internet search should help you resolve the matter quickly.

Linux Shell

The default login shell for your user account is Bash. To determine your current login shell, execute:

$ echo $SHELL

If you'd like to change your login shell to csh, sh, tcsh, or zsh, submit a ticket through the TACC or XSEDE portal. The "chsh" ("change shell") command will not work on TACC systems.

When you start a shell on Stampede2, system-level startup files initialize your account-level environment and aliases before the system sources your own user-level startup scripts. You can use these startup scripts to customize your shell by defining your own environment variables, aliases, and functions. These scripts (e.g. .profile and .bashrc) are generally hidden files: so-called dotfiles that begin with a period, visible when you execute: "ls -a".

Before editing your startup files, however, it's worth taking the time to understand the basics of how your shell manages startup. Bash startup behavior is very different from the simpler csh behavior, for example. The Bash startup sequence varies depending on how you start the shell (e.g. using ssh to open a login shell, execute the "bash" command to begin an interactive shell, or launching a script to start a non-interactive shell). Moreover, Bash does not automatically source your .bashrc when you start a login shell by using ssh to connect to a node. Unless you have specialized needs, however, this is undoubtedly more flexibility than you want: you will probably want your environment to be the same regardless of how you start the shell. The easiest way to achieve this is to execute "source ~/.bashrc" from your ".profile", then put all your customizations in ".bashrc". The system-generated default startup scripts demonstrate this approach. We recommend that you use these default files as templates.

For more information see the Bash Users' Startup Files: Quick Start Guide and other online resources that explain shell startup. To recover the originals that appear in a newly created account, execute "/usr/local/startup_scripts/install_default_scripts".

Environment Variables

Your environment includes the environment variables and functions defined in your current shell: those initialized by the system, those you define or modify in your account-level startup scripts, and those defined or modified by the modules that you load to configure your software environment. Be sure to distinguish between an environment variable's name (e.g. HISTSIZE) and its value ($HISTSIZE). Understand as well that a sub-shell (e.g. a script) inherits environment variables from its parent, but does not inherit ordinary shell variables or aliases. Use export (in Bash) or setenv (in csh) to define an environment variable.

Execute the "env" command to see the environment variables that that define the way your shell and child shells behave.

Pipe the results of env into grep to focus on specific environment variables. For example, to see all environment variables that contain the string GIT (in all caps), execute:

$ env | grep GIT

The environment variables PATH and LD_LIBRARY_PATH are especially important. PATH is a colon-separated list of directory paths that determines where the system looks for your executables. LD_LIBRARY_PATH is a similar list that determines where the system looks for shared libraries.

Account-Level Diagnostics

TACC's sanitytool module loads an account-level diagnostic package that detects common account-level issues and often walks you through the fixes. You should certainly run the package's sanitycheck utility when you encounter unexpected behavior. You may also want to run sanitycheck periodically as preventive maintenance. To run sanitytool's account-level diagnostics, execute the following commands:

login1$ module load sanitytool
login1$ sanitycheck

Execute "module help sanitytool" for more information.

Accessing the Compute Nodes

You connect to Stampede2 through one of four "front-end" login nodes. The login nodes are shared resources: at any given time there are many users logged into each of these login nodes, each preparing to access the "back-end" compute nodes (Figure 2. Login and Compute Nodes). What you do on the login nodes affects other users directly because you are competing for the same memory and processing power. This is the reason you should not run your applications on the login nodes or otherwise abuse them. Think of the login nodes as a prep area where you can manage files and compile code before accessing the compute nodes to perform research computations. See Good Citizenship for more information.

You can use your command-line prompt, or the "hostname" command, to tell you whether you are on a login node or a compute node. The default prompt, or any custom prompt containing "\h", displays the short form of the hostname (e.g. c401-064). The hostname for a Stampede2 login node begins with the string "login" (e.g. login2.stampede2.tacc.utexas.edu), while compute node hostnames begin with the character "c" (e.g. c401-064.stampede2.tacc.utexas.edu). Note that the default prompts on the compute nodes include the node type (knl or skx) as well. The environment variable TACC_NODE_TYPE, defined only on the compute nodes, also displays the node type. The simplified prompts in the User Guide examples are shorter than Stampede2's actual default prompts.

While some workflows, tools, and applications hide the details, there are three basic ways to access the compute nodes:

  1. Submit a batch job using the sbatch command. This directs the scheduler to run the job unattended when there are resources available. Until your batch job begins it will wait in a queue. You do not need to remain connected while the job is waiting or executing. See Running Jobs for more information. Note that the scheduler does not start jobs on a first come, first served basis; it juggles many variables to keep the machine busy while balancing the competing needs of all users. The best way to minimize wait time is to request only the resources you really need: the scheduler will have an easier time finding a slot for the two hours you need than for the 48 hours you unnecessarily request.
  2. Begin an interactive session using idev or srun. This will log you into a compute node and give you a command prompt there, where you can issue commands and run code as if you were doing so on your personal machine. An interactive session is a great way to develop, test, and debug code. When you request an interactive session, the scheduler submits a job on your behalf. You will need to remain logged in until the interactive session begins.
  3. Begin an interactive session using ssh to connect to a compute node on which you are already running a job. This is a good way to open a second window into a node so that you can monitor a job while it runs.

Be sure to request computing resources that are consistent with the type of application(s) you are running:

  • A serial (non-parallel) application can only make use of a single core on a single node, and will only see that node's memory.
  • A threaded program (e.g. one that uses OpenMP) employs a shared memory programming model and is also restricted to a single node, but the program's individual threads can run on multiple cores on that node.
  • An MPI (Message Passing Interface) program can exploit the distributed computing power of multiple nodes: it launches multiple copies of its executable (MPI tasks, each assigned unique IDs called ranks) that can communicate with each other across the network. The tasks on a given node, however, can only directly access the memory on that node. Depending on the program's memory requirements, it may not be possible to run a task on every core of every node assigned to your job. If it appears that your MPI job is running out of memory, try launching it with fewer tasks per node to increase the amount of memory available to individual tasks.
  • A popular type of parameter sweep (sometimes called high throughput computing) involves submitting a job that simultaneously runs many copies of one serial or threaded application, each with its own input parameters ("Single Program Multiple Data", or SPMD). The "launcher" tool is designed to make it easy to submit this type of job. For more information:

    $ module load launcher
    $ module help launcher

Figure 2. Login and compute nodes

Using Modules to Manage your Environment

Lmod, a module system developed and maintained at TACC, makes it easy to manage your environment so you have access to the software packages and versions that you need to conduct your research. This is especially important on a system like Stampede2 that serves thousands of users with an enormous range of needs. Loading a module amounts to choosing a specific package from among available alternatives:

$ module load intel          # load the default Intel compiler
$ module load intel/17.0.4   # load a specific version of Intel compiler

A module does its job by defining or modifying environment variables (and sometimes aliases and functions). For example, a module may prepend appropriate paths to $PATH and $LD_LIBRARY_PATH so that you can find the executables and libraries associated with a given software package. The module creates the illusion that the system is installing software for your personal use. Unloading a module reverses these changes and creates the illusion that the system just uninstalled the software:

$ module load   ddt  # defines DDT-related env vars; modifies others
$ module unload ddt  # undoes changes made by load

The module system does more, however. When you load a given module, the module system can automatically replace or deactivate modules to ensure the packages you have loaded are compatible with each other. In the example below, the module system automatically unloads one compiler when you load another, and replaces Intel-compatible versions of IMPI and PETSc with versions compatible with gcc:

$ module load intel  # load default version of Intel compiler
$ module load petsc  # load default version of PETSc
$ module load gcc    # change compiler

Lmod is automatically replacing "intel/17.0.4" with "gcc/7.1.0".

Due to MODULEPATH changes, the following have been reloaded:
1) impi/17.0.3     2) petsc/3.7

On Stampede2, modules generally adhere to a TACC naming convention when defining environment variables that are helpful for building and running software. For example, the "papi" module defines TACC_PAPI_BIN (the path to PAPI executables), TACC_PAPI_LIB (the path to PAPI libraries), TACC_PAPI_INC (the path to PAPI include files), and TACC_PAPI_DIR (top-level PAPI directory). After loading a module, here are some easy ways to observe its effects:

$ module show papi   # see what this module does to your environment
$ env | grep PAPI    # see env vars that contain the string PAPI
$ env | grep -i papi # case-insensitive search for 'papi' in environment

To see the modules you currently have loaded:

$ module list

To see all modules that you can load right now because they are compatible with the currently loaded modules:

$ module avail

To see all installed modules, even if they are not currently available because they are incompatible with your currently loaded modules:

$ module spider   # list all modules, even those not available to load

To filter your search:

$ module spider slep             # all modules with names containing 'slep'
$ module spider sundials/2.5.0   # additional details on a specific module

Among other things, the latter command will tell you which modules you need to load before the module is available to load. You might also search for modules that are tagged with a keyword related to your needs (though your success here depends on the diligence of the module writers). For example:

$ module keyword performance

You can save a collection of modules as a personal default collection that will load every time you log into Stampede2. To do so, load the modules you want in your collection, then execute:

$ module save    # save the currently loaded collection of modules 

Two commands make it easy to return to a known, reproducible state:

$ module reset   # load the system default collection of modules
$ module restore # load your personal default collection of modules

On TACC systems, the command "module reset" is equivalent to "module purge; module load TACC". It's a safer, easier way to get to a known baseline state than issuing the two commands separately.

Help text is available for both individual modules and the module system itself:

$ module help swr     # show help text for software package swr
$ module help         # show help text for the module system itself

See Lmod's online documentation for more extensive documentation. The online documentation addresses several topics (e.g. writing and using your own module files) that are beyond the scope of the help text.

It's safe to execute module commands in job scripts. In fact this is a good way to write self-documenting, portable job scripts that produce reproducible results. If you use "module save" to define a personal default module collection, it's rarely necessary to execute module commands in shell startup scripts, and it can be tricky to do so safely. If you do wish to put module commands in your startup scripts, see Stampede2's default startup scripts for a safe way to do so.

Good Citizenship

You share Stampede2 with thousands of other users, and what you do on the system affects others. Exercise good citizenship to ensure that your activity does not adversely impact the system and the research community with whom you share it. Here are some rules of thumb.

Login Nodes

When you connect to Stampede2 you share the login node with dozens of other users.

  • Know when you're running on a login node. You can use your Linux prompt, the "hostname" command, or other mechanisms to do so. See Accessing the Compute Nodes for more information.

  • Know what's appropriate on a login node. A login node is a good place to edit and manage files, initiate file transfers, compile code, submit new jobs, and track existing jobs.

  • Avoid computationally intensive activity on login nodes. This means:

    • Don't run research applications on the login nodes; this includes frameworks like MATLAB and R.
    • Don't launch too many simultaneous processes: while it's fine to compile on a login node, "make -j 16" (which compiles on 16 cores) may be a bit rude.
    • That script you wrote to check job status should probably do so every few minutes rather than several times a second.

Shared Lustre File Systems

This section focuses on ways to avoid causing problems on $HOME, $WORK, and $SCRATCH. File Systems above is a brief overview of these file systems. Configuring Your Account covers environment variables and aliases that help you navigate the file systems. File Operations addresses optimization and parallel I/O.

  • Stripe the receiving directory before creating large files in the directory or transferring large files to the directory. See Striping Large Files) for more information.

  • Don't run jobs in $HOME. The $HOME file system is for routine file management, not parallel jobs.

  • Run I/O intensive jobs in $SCRATCH rather than $WORK. If you stress $WORK, you affect every user on every TACC system.

  • Don't get greedy. If you know or suspect your workflow is I/O intensive, don't submit a pile of simultaneous jobs. Writing restart/snapshot files can stress the file system; avoid doing so too frequently.

  • Watch your file system quotas. If you're near your quota in $WORK and your job is repeatedly trying (and failing) to write to $WORK, you will stress the file system. If you're near your quota in $HOME, jobs run on any file system may fail, because all jobs write some data to the hidden $HOME/.slurm directory.

  • Avoid opening and closing files repeatedly in tight loops. Every open/close operation requires the MDS, which is a single point of failure. If possible, open files once at the beginning of your program/workflow, then close them at the end.

  • Avoid storing many small files in a single directory, and avoid workflows that require many small files. A few hundred files in a single directory is probably fine; tens of thousands is almost certainly too many. If you must use many small files, group them in separate directories of manageable size.

Internal and External Networks

  • Avoid too many simultaneous file transfers. You share the network bandwidth with other users; don't use more than your fair share. Two or three concurrent scp sessions is probably fine. Twenty is probably not.

  • Avoid recursive file transfers, especially those involving many small files. Create a tar archive before transfers. This is especially true when transferring files to or from Ranch.

Submitting Jobs

  • When you submit a job to the scheduler, don't ask for more time than you really need. The scheduler will have an easier time finding you a slot for the 2 hours you need than the 48 hours you request. This means shorter queue waits times for you and everybody else.

  • Test your submission scripts. Start small: make sure everything works on 2 nodes before you try 200. Work out submission bugs and kinks with 5 minute jobs that won't wait long in the queue and involve short, simple substitutes for your real workload: simple test problems; "hello world" codes; one-liners like "ibrun hostname"; or an ldd on your executable.

  • Respect memory limits and other system constraints. If your application needs more memory than is available, your job will fail, and may leave nodes in unusable states. Monitor your application's needs. Execute "module load remora" followed by "module help remora" for more information on a particularly handy monitoring tool.

Help Desk Tickets

  • Do your homework before submitting a help desk ticket. What does the user guide and other documentation say? Search the internet for key phrases in your error logs; that's probably what the consultants answering your ticket are going to do. What have you changed since the last time your job succeeded?

  • Subscribe to TACC and/or XSEDE User News. This is the best way to keep abreast of maintenance schedules, system outages, and other general interest items.

  • Have realistic expectations. Consultants can address system issues and answer questions about Stampede2. But they can't teach parallel programming in a ticket, and may know nothing about the package you downloaded. They may offer general advice that will help you build, debug, optimize, or modify your code, but you shouldn't expect them to do these things for you.

  • Ask your question once in a single ticket. Note that tickets submitted through the TACC and XSEDE user portals are routed to the same consultants.

  • Be patient. It may take a business day for the consultant to get back to you, especially if your issue is complex. It might take an exchange or two before you and the consultant are on the same page. If the admins disable your account, it's not punitive. When the file system is in danger of crashing, or a login node hangs, they don't have time to notify you before taking action.

Stampede2 mounts three file Lustre file systems that are shared across all nodes: the home, work, and scratch file systems. Stampede2's startup mechanisms define corresponding account-level environment variables $HOME, $SCRATCH, and $WORK that store the paths to directories that you own on each of these file systems. Consult the Stampede2 File Systems table for the basic characteristics of these file systems, File Operations for advice on performance issues, and Good Citizenship for file-related tips on good citizenship.

The system defines several account-level aliases that make it easy to navigate across the directories you own in these file systems:

Table 4. Built-in Account Level Aliases

Built-in Account Level Aliases
Alias Command
cd or cdh cd $HOME
cdw cd $WORK
cds cd $SCRATCH

Stampede2's home and scratch file systems are mounted only on Stampede2, but the work file system mounted on Stampede2 is the Global Shared File System hosted on Stockyard. It is the same file system that is available on Stampede1, Maverick, Wrangler, Lonestar 5, and other TACC resources.

The $STOCKYARD environment variable points to the highest-level directory that you own on the Global Shared File System. The definition of the $STOCKYARD environment variable is of course account-specific, but you will see the same value on all TACC systems (See Figure 3. Stockyard). This directory is an excellent place to store files you want to access regularly from multiple TACC resources.

Your account-specific $WORK environment variable varies from system to system and (except for Stampede1) is a sub-directory of $STOCKYARD. The sub-directory name corresponds to the associated TACC resource. The $WORK environment variable on Stampede2 points to the $STOCKYARD/stampede2 subdirectory, a convenient location for files you use and jobs you run on Stampede2. Remember, however, that all subdirectories contained in your $STOCKYARD directory are available to you from any system that mounts the file system. If you have accounts on both Stampede2 and Maverick, for example, the $STOCKYARD/stampede2 directory is available from your Maverick account, and $STOCKYARD/maverick is available from your Stampede2 account. Your quota and reported usage on the Global Shared File System reflects all files that you own on Stockyard, regardless of their actual location on the file system.

Note that resource-specific sub-directories of $STOCKYARD are nothing more than convenient ways to manage your resource-specific files. You have access to any such sub-directory from any TACC resources. If you are logged into Stampede2, for example, executing the alias cdw (equivalent to "cd $WORK") will take you to the resource-specific sub-directory $STOCKYARD/stampede2. But you can access this directory from other TACC systems by executing "cd $STOCKYARD/stampede2". This makes it particularly easy to share files across TACC systems.

$WORK: Stampede2 vs Stampede1

Stampede2 defines the $WORK environment variable differently than Stampede1 did: your Stampede2 $WORK directory is a sub-directory of your Stampede1 work directory. On Stampede2, your $WORK directory is $STOCKYARD/stampede2 (e.g. /work/01234/bjones/stampede2). On Stampede1, your $WORK directory was the $STOCKYARD directory itself (e.g. /work/01234/bjones).

Please see an example for fictitious user bjones in the figure below. All directories are accessible from all systems. A given sub-directory (e.g. wrangler, maverick) exists only if you have an allocation on the respective system.

Figure 3. Account-level directories on the work file system (Global Shared File System hosted on Stockyard).

Temporary Mounts of Stampede1 File Systems

For your convenience during the transition from Stampede1 to Stampede2, the Stampede1 home and scratch file systems are available as read-only Lustre file systems on the Stampede2 login nodes (and only the login nodes). The mount points on the Stampede2 logins are /oldhome1 and /oldscratch respectively, and your account includes the environment variables $OLDHOME and $OLDSCRATCH pointing to your Stampede1 $HOME and $SCRATCH directories respectively. The aliases "cdoh" and "cdos", defined in your Stampede2 account, are equivalent to "cd $OLDHOME" and "cd $OLDSCRATCH" (see the Built-In Account-level Aliases table above).

Do not submit Stampede2 jobs (sbatch, srun, or idev) from directories in $OLDHOME or $OLDSCRATCH. Because these directories are read-only, attempting to do so may lead to job failures or other subtle problems that may prove difficult to diagnose.

Your Stampede1 $WORK directory is, of course, available to you from Stampede2 (see Monitoring Jobs and Queues above). As a matter of convenience, however, your Stampede2 account includes the environment variable $OLDWORK (which has the same value as $STOCKYARD) and the associated alias cdow.

Transferring Files from Stampede1 to Stampede2

Transfers from $OLDHOME and $OLDSCRATCH: Stampede2's temporary mounts of the Stampede1 file system make it easy to transfer files. For example, the following command:

login1$ cp -r $OLDHOME/mysrc $HOME

copies the directory mysrc and its contents from your Stampede1 home directory to your Stampede2 home directory. The rsync command is also available. Given the temporary mounts, there's little reason to use scp. In any case, please remember that recursive copy operations can put a significant strain on Lustre file systems: copy only the files you need, and don't execute more than one or two simultaneous recursive copies.

Transfers involving $WORK: When transferring files on the Stockyard-hosted work file system, it's probably best to use mv rather than cp. If you use cp you will end up with two copies of your file(s) on the work file system; at the very least, this will put pressure on your file system quota. The mv command will not work when transferring files from $OLDHOME or $OLDSCRATCH because these two file systems are read-only on Stampede2.

Striping Large Files: Before copying large files to Stampede2 be sure to set an appropriate default stripe count on the receiving directory. See "Striping Large Files" for more information.

Startup Files: It is generally safe to copy your startup files (e.g. .profile and .bashrc) from Stampede1 to Stampede2, though you may of course have to make some changes. Execute "/usr/local/startup_scripts/install_default_scripts" to recover the default startup files that appear in a newly created account.

Transferring Files Using scp and rsync

You can transfer files between Stampede2 and Linux-based systems using either scp or rsync. Both scp and rsync are available in the Mac Terminal app. Windows ssh clients typically include scp-based file transfer capabilities.

The Linux scp (secure copy) utility is a component of the OpenSSH suite. Assuming your Stampede2 username is bjones, a simple scp transfer that pushes a file named "myfile" from your local Linux system to Stampede2 $HOME would look like this:

localhost$ scp ./myfile bjones@stampede2.tacc.utexas.edu:  # note colon after net address

To pull myfile from Stampede2 $HOME to the current directory on your local system (represented by "." in the command below):

localhost$ scp bjones@stampede2.tacc.utexas.edu:myfile .   # note colon after net address

To pull all files ending in ".sh" from /work/01234/bjones/scripts on Stampede2 to the current directory on your local system:

localhost$ scp bjones@stampede2.tacc.utexas.edu:/work/01234/bjones/stampede2/*.sh .  

You can of course use shell or environment variables in your calls to scp. For example:

localhost$ destdir="/work/01234/bjones/stampede2/data"
localhost$ scp ./myfile bjones@stampede2.tacc.utexas.edu:$destdir

You can also issue scp commands on your local client that use Stampede2 environment variables like $HOME, $WORK, and $SCRATCH. To do so, use a backslash ("\") as an escape character before the "$"; this ensures that expansion occurs after establishing the connection to Stampede2:

localhost$ scp ./myfile bjones@stampede2.tacc.utexas.edu:\$WORK/data   # Note backslash

Avoid using scp for recursive ("-r") transfers of directories that contain nested directories of many small files:

localhost$ scp -r  ./mydata     bjones@stampede2.tacc.utexas.edu:\$WORK  # DON'T DO THIS

Instead, use tar to create an archive of the directory, then transfer the directory as a single file:

localhost$ tar cvf ./mydata.tar mydata                                   # create archive
localhost$ scp     ./mydata.tar bjones@stampede2.tacc.utexas.edu:\$WORK  # transfer archive

The rsync (remote synchronization) utility is a great way to synchronize files that you maintain on more than one system: when you transfer files using rsync, the utility copies only the changed portions of individual files. As a result, rsync is especially efficient when you only need to update a small fraction of a large dataset. The basic syntax is similar to scp:

localhost$ rsync       mybigfile bjones@stampede2.tacc.utexas.edu:\$WORK/data
localhost$ rsync -avtr mybigdir  bjones@stampede2.tacc.utexas.edu:\$WORK/data

The options on the second transfer are typical and appropriate when synching a directory: this is a recursive update ("-r") with verbose ("-v") feedback; the synchronization preserves time stamps ("-t") as well as symbolic links and other meta-data ("-a"). Because rsync only transfers changes, recursive updates with rsync may be less demanding than an equivalent recursive transfer with scp.

See Good Citizenship for additional important advice about striping the receiving directory when transferring large files; watching your quota on $HOME and $WORK; and limiting the number of simultaneous transfers. Remember also that $STOCKYARD (and your $WORK directory on each TACC resource) is available from all major TACC systems: there's no need for scp when both the source and destination involve sub-directories of $STOCKYARD. See Managing Your Files for more information about transfers on $STOCKYARD.

Transferring Files Using Globus

Globus is an excellent way to transfer data between XSEDE sites. It provides fast, secure transfers based on the GridFTP protocol, and has an easy-to-use web interface that lets you move data between pre-defined as well as customized "endpoints". XSEDE users may access Globus using their XSEDE User Portal credentials. You can add a customized local endpoint (e.g. a personal computer) after signing up for a free Globus Connect Personal account.

globus-url-copy

XSEDE users may also use Globus' globus-url-copy command-line utility to transfer data between XSEDE sites. globus-url-copy, like Globus Connect described above, is an implementation of the GridFTP protocol, providing high speed transport between GridFTP servers at XSEDE sites. The GridFTP servers mount the specific file systems of the target machine, thereby providing access to your files or directories.

This command requires the use of an XSEDE certificate to create a proxy for passwordless transfers. To obtain a proxy, use the "myproxy-logon" command with your XSEDE User Portal (XUP) username and password to obtain a proxy certificate. The proxy is valid for 12 hours for all logins on the local machine. On Stampede, the myproxy-logon command is located in the CTSSV4 module (not loaded by default).

login1$ module load CTSSV4
login1$ myproxy-logon -T -l XUP_username

Each globus-url-copy invocation must include the name of the server and a full path to the file. The general syntax looks like:

globus-url-copy [options] source_url destination_url

where each XSEDE URL will generally be formatted:

gsiftp://gridftp_server/path/to/file

Note that globus-url-copy supports multiple protocols e.g., HTTP, FTP in addition to the GridFTP protocol. Please consult the following references for more information.

globus-url-copy Examples

The following command copies "directory1" from TACC's Stampede2 to PSC Data Supercell system home file system, renaming it to "directory2". Note that when transferring directories, the directory path must end with a slash ( "/"):

login1$ globus-url-copy -r -vb \ 
    gsiftp://gridftp.stampede2.tacc.xsede.org:2811/`pwd`/directory1/ \ 
    gsiftp://gridftp.psc.xsede.org:2811/~/directory2/

The mapping of "/~/" depends on the configuration of the GridFTP server but is typically the local user's home directory on Linux systems.

The following command copies a single file, "file1" from TACC's Stampede2 to "file2" on Stanford's XStream home file system:

login1$ globus-url-copy -tcp-bs 11M -vb \ 
    gsiftp://gridftp.stampede2.tacc.xsede.org:2811/`pwd`/file1 \ 
    gsiftp://xstream.stanford.xsede.org:2811/~/file2

Use the buffer size option, "-tcp-bs 11M", to explicitly set the FTP data channel buffer size, otherwise, the speed will be about 20 times slower! Consult the Globus documentation to select the optimum value: How do I choose a value for the TCP buffer size (-tcp) option?

Advanced users may employ the "-stripe" option enables striped transfers on supported servers. Stampede's GridFTP servers each have a 10GbE interface adapter and are configured for a 4-way stripe since most deployed 10GbE interfaces are performance-limited by host PCI-X busses to ~6Gb/s.

Sharing Files with Collaborators

If you wish to share files and data with collaborators in your project, see Sharing Project Files on TACC Systems for step-by-step instructions. Project managers or delegates can use Unix group permissions and commands to create read-only or read-write shared workspaces that function as data repositories and provide a common work area to all project members.

Striping Large Files

Before transferring large files to Stampede2, or creating new large files, be sure to set an appropriate default stripe count on the receiving directory. To avoid exceeding your fair share of any given OST, a good rule of thumb is to allow at least one stripe for each 100GB in the file. For example, to set the default stripe count on the current directory to 30 (a plausible stripe count for a directory receiving a file approaching 3TB in size), execute:

$ lfs setstripe -c 30 $PWD

Note that an "lfs setstripe" command always sets both stripe count and stripe size, even if you explicitly specify only one or the other. Since the example above does not explicitly specify stripe size, the command will set the stripe size on the directory to Stampede2's system default (1MB). In general there's no need to customize stripe size when creating or transferring files.

Building Software

The phrase "building software" is a common way to describe the process of producing a machine-readable executable file from source files written in C, Fortran, or some other programming language. In its simplest form, building software involves a simple, one-line call or short shell script that invokes a compiler. More typically, the process leverages the power of makefiles, so you can change a line or two in the source code, then rebuild in a systematic way only the components affected by the change. Increasingly, however, the build process is a sophisticated multi-step automated workflow managed by a special framework like autotools or cmake, intended to achieve a repeatable, maintainable, portable mechanism for installing software across a wide range of target platforms.

Basics of Building Software

This section of the user guide does nothing more than introduce the big ideas with simple one-line examples. You will undoubtedly want to explore these concepts more deeply using online resources. You will quickly outgrow the examples here. We recommend that you master the basics of makefiles as quickly as possible: even the simplest computational research project will benefit enormously from the power and flexibility of a makefile-based build process.

Intel Compilers

Intel is the recommended and default compiler suite on Stampede2. Each Intel module also gives you direct access to mkl without loading an mkl module; see Intel MKL for more information. Here are simple examples that use the Intel compiler to build an executable from source code:

$ icc mycode.c                    # C source file; executable a.out
$ icc main.c calc.c analyze.c     # multiple source files
$ icc mycode.c     -o myexe       # C source file; executable myexe
$ icpc mycode.cpp  -o myexe       # C++ source file
$ ifort mycode.f90 -o myexe       # Fortran90 source file

Compiling a code that uses OpenMP would look like this:

$ icc -qopenmp mycode.c -o myexe  # OpenMP

See the published Intel documentation, available both online and in ${TACC_INTEL_DIR}/documentation, for information on optimization flags and other Intel compiler options.

GNU Compilers

The GNU foundation maintains a number of high quality compilers, including a compiler for C (gcc), C++ (g++), and Fortran (gfortran). The gcc compiler is the foundation underneath all three, and the term "gcc" often means the suite of these three GNU compilers.

Load a gcc module to access a recent version of the GNU compiler suite. Avoid using the GNU compilers that are available without a gcc module - those will be older versions based on the "system gcc" that comes as part of the Linux distribution.

Here are simple examples that use the GNU compilers to produce an executable from source code:

$ gcc mycode.c                    # C source file; executable a.out
$ gcc mycode.c          -o myexe  # C source file; executable myexe
$ g++ mycode.cpp        -o myexe  # C++ source file
$ gfortran mycode.f90   -o myexe  # Fortran90 source file
$ gcc -fopenmp mycode.c -o myexe  # OpenMP; GNU flag is different than Intel

Note that some compiler options are the same for both Intel and GNU (e.g. "-o"), while others are different (e.g. "-qopenmp" vs "-fopenmp"). Many options are available in one compiler suite but not the other. See the online GNU documentation for information on optimization flags and other GNU compiler options.

Include and Library Paths

Software often depends on pre-compiled binaries called libraries. When this is true, compiling usually requires using the "-I" option to specify paths to so-called header or include files that define interfaces to the procedures and data in those libraries. Similarly, linking often requires using the "-L" option to specify paths to the libraries themselves. Typical compile and link lines might look like this:

$ icc        -c main.c -I${WORK}/mylib/inc -I${TACC_HDF5_INC}                  # compile
$ icc main.o -o myexe  -L${WORK}/mylib/lib -L${TACC_HDF5_LIB} -lmylib -lhdf5   # link

On Stampede2, both the hdf5 and phdf5 modules define the environment variables $TACC_HDF5_INC and $TACC_HDF5_LIB. Other module files define similar environment variables; see Using Modules for more information.

The details of the linking process vary, and order sometimes matters. Much depends on the type of library: static (.a suffix; library's binary code becomes part of executable image at link time) versus dynamically-linked shared (.so suffix; library's binary code is not part of executable; it's located and loaded into memory at run time). The link line can use rpath to store in the executable an explicit path to a shared library. In general, however, the LD_LIBRARY_PATH environment variable specifies the search path for dynamic libraries. For software installed at the system-level, TACC's modules generally modify LD_LIBRARY_PATH automatically. To see whether and how an executable named "myexe" resolves dependencies on dynamically linked libraries, execute "ldd myexe".

A separate section below addresses the Intel Math Kernel Library (MKL).

Compiling and Linking MPI Programs

Intel MPI (module impi) is available on Stampede2 for both the Intel and GNU compilers. After loading an impi module, compile and/or link by using an mpi wrapper (mpicc, mpicxx, mpif90) in place of the compiler:

$ mpicc    mycode.c   -o myexe   # C source, full build
$ mpicc -c mycode.c              # C source, compile without linking
$ mpicxx   mycode.cpp -o myexe   # C++ source, full build
$ mpif90   mycode.f90 -o myexe   # Fortran source, full build

These wrappers call the compiler with the options, include paths, and libraries necessary to produce an MPI executable using the MPI module you're using. To see the effect of a given wrapper, call it with the "-show" option:

$ mpicc -show  # Show compile line generated by call to mpicc; similarly for other wrappers

Building Third-Party Software in Your Own Account

You're welcome to download third-party research software and install it in your own account. In most cases you'll want to download the source code and build the software so it's compatible with the Stampede2 software environment. You can't use yum or any other installation process that requires elevated privileges, but this is almost never necessary. The key is to specify an installation directory for which you have write permissions. Details vary; you should consult the package's documentation and be prepared to experiment. When using the famous three-step autotools build process, the standard approach is to use the PREFIX environment variable to specify a non-default, user-owned installation directory at the time you execute configure or make:

$ export PREFIX=$WORK/apps/t3pio
$ ./configure --prefix=$PREFIX
$ make
$ make install

Other languages, frameworks, and build systems generally have equivalent mechanisms for installing software in user space. In most cases a web search like "Python Linux install local" will get you the information you need.

In Python, a local install looks like this:

$ pip install             --user netCDF4     # install to $HOME/.local
$ python setup.py install --user             # install to $HOME/.local
$ pip install --install-option=$INSTALLDIR   # install to $INSTALLDIR

Similarly in R:

$ export R_LIBS="$WORK/myR"     # defines default install dir for install.packages
$ install.packages(devtools)    # install to location specified by R_LIBS

You may, of course, need to customize the build process in other ways. It's likely, for example, that you'll need to edit a makefile or other build artifact to specify Stampede2-specific include and library paths or other compiler settings. A good way to proceed is to write a shell script that implements the entire process: definitions of environment variables, module commands, and calls to the build utilities. Include echo statements with appropriate diagnostics. Run the script until you encounter an error. Research and fix the current problem. Document your experience in the script itself; including dead-ends, alternatives, and lessons learned. Re-run the script to get to the next error, then repeat until done. When you're finished, you'll have a repeatable process that you can archive until it's time to update the software or move to a new machine.

Intel Math Kernel Library (MKL)

The Intel Math Kernel Library (MKL) is a collection of highly optimized functions implementing some of the most important mathematical kernels used in computational science, including standardized interfaces to:

  • BLAS (Basic Linear Algebra Subroutines), a collection of low-level matrix and vector operations like matrix-matrix multiplication
  • LAPACK (Linear Algebra PACKage), which includes higher-level linear algebra algorithms like Gaussian Elimination
  • FFT (Fast Fourier Transform), including interfaces based on FFTW (Fastest Fourier Transform in the West)
  • ScaLAPACK (Scalable LAPACK), BLACS (Basic Linear Algebra Communication Subprograms), Cluster FFT, and other functionality that provide block-based distributed memory (multi-node) versions of selected LAPACK, BLAS, and FFT algorithms;
  • Vector Mathematics (VM) functions that implement highly optimized and vectorized versions of special functions like sine and square root.

MKL with Intel C, C++, and Fortran Compilers

There is no MKL module for the Intel compilers because you don't need one: the Intel compilers have built-in support for MKL. Unless you have specialized needs, there is no need to specify include paths and libraries explicitly. Instead, using MKL with the Intel modules requires nothing more than compiling and linking with the "-mkl" option.; e.g.

$ icc   -mkl mycode.c
$ ifort -mkl mycode.c

The "-mkl" switch is an abbreviated form of "-mkl=parallel", which links your code to the threaded version of MKL. To link to the unthreaded version, use "-mkl=sequential". A third option, "-mkl=cluster", which also links to the unthreaded libraries, is necessary and appropriate only when using ScaLAPACK or other distributed memory packages. For additional information, including advanced linking options, see the MKL documentation and Intel MKL Link Line Advisor.

MKL with GNU C, C++, and Fortran Compilers

When using a GNU compiler, load the MKL module before compiling or running your code, then specify explicitly the MKL libraries, library paths, and include paths your application needs. Consult the Intel MKL Link Line Advisor for details. A typical compile/link process on a TACC system will look like this:

$ module load gcc
$ module load mkl                         # available/needed only for GNU compilers
$ gcc -fopenmp -I$MKLROOT/include         \
         -Wl,-L${MKLROOT}/lib/intel64     \
         -lmkl_intel_lp64 -lmkl_core      \
         -lmkl_gnu_thread -lpthread       \
         -lm -ldl mycode.c

For your convenience the mkl module file also provides alternative TACC-defined variables like $TACC_MKL_INCLUDE (equivalent to $MKLROOT/include). Execute "module help mkl" for more information.

Using MKL as BLAS/LAPACK with Third-Party Software

When your third-party software requires BLAS or LAPACK, you can use MKL to supply this functionality. Replace generic instructions that include link options like "-lblas" or "-llapack" with the simpler MKL approach described above. There is no need to download and install alternatives like OpenBLAS.

Using MKL as BLAS/LAPACK with TACC's MATLAB, Python, and R Modules

TACC's MATLAB, Python, and R modules all use threaded (parallel) MKL as their underlying BLAS/LAPACK library. These means that even serial codes written in MATLAB, Python, or R may benefit from MKL's thread-based parallelism. This requires no action on your part other than specifying an appropriate max thread count for MKL; see the section below for more information.

Controlling Threading in MKL

Any code that calls MKL functions can potentially benefit from MKL's thread-based parallelism; this is true even if your code is not otherwise a parallel application. If you are linking to the threaded MKL (using "-mkl", "-mkl=parallel", or the equivalent explicit link line), you need only specify an appropriate value for the max number of threads available to MKL. You can do this with either of the two environment variables MKL_NUM_THREADS or OMP_NUM_THREADS. The environment variable MKL_NUM_THREADS specifies the max number of threads available to each instance of MKL, and has no effect on non-MKL code. If MKL_NUM_THREADS is undefined, MKL uses OMP_NUM_THREADS to determine the max number of threads available to MKL functions. In either case, MKL will attempt to choose an optimal thread count less than or equal to the specified value. Note that OMP_NUM_THREADS defaults to 1 on TACC systems; if you use the default value you will get no thread-based parallelism from MKL.

If you are running a single serial, unthreaded application (or an unthreaded MPI code involving a single MPI task per node) it is usually best to give MKL as much flexibility as possible by setting the max thread count to the total number of hardware threads on the node (272 on KNL). Of course things are more complicated if you are running more than one process on a node: e.g. multiple serial processes, threaded applications, hybrid MPI-threaded applications, or pure MPI codes running more than one MPI rank per node. See http://software.intel.com/en-us/articles/recommended-settings-for-calling-intel-mkl-routines-from-multi-threaded-applications and related Intel resources for examples of how to manage threading when calling MKL from multiple processes.

Building Software for Phase 1 Compute Nodes (KNL)

You can compile for the Stampede2 KNLs on either a Broadwell login node or any KNL compute node. Building on the login node is likely to be faster, and is the approach we currently recommend. In either case, use the "-xMIC-AVX512" switch at both compile and link time to produce compiled code targeting the KNL. In addition, you may want to specify an optimization level (e.g. "-O3"). You may want to avoid using "-xHost" when building on a Broadwell login node. If you do so you will produce an executable that will run on KNL but will not employ its optimized instruction set.

When building on a login node using build systems that compile and run their own test programs (e.g. Autotools/configure, SCons, and Cmake), you will need to specify flags that produce code that will run on both the Broadwell login node (the build architecture where these tests will run) and on the compute KNL nodes (the actual target architecture). This is done through an Intel compiler feature called CPU dispatch that produces binaries containing alternate paths with optimized codes for multiple architectures. To produce such a binary containing optimized code for both Broadwell and KNL, supply two flags when compiling and linking (the same settings you would use when building on the Haswell login node that was the front end for the Stampede1 KNL sub-system):

-xCORE-AVX2 -axMIC-AVX512

In a typical build system, add these flags to the CFLAGS, CXXFLAGS, FFLAGS, and LDFLAGS variables. Expect the build to take longer than it would for one target architecture, and expect the resulting binary to be larger.

Stampede2's Intel compilers are newer than those that were installed on the Stampede1 KNL sub system. We therefore recommend rebuilding software originally compiled with Intel for the Stampede1 KNL sub-system.

Building Software for Phase 2 Compute Nodes (SKX)

Pending. This material will be available during the Phase 2 SKX early user period.

Job Accounting

Stampede2's accounting system is based on node-hours: one Service Unit (SU) represents a single compute node used for one hour (a node-hour). The total cost of any given job is total core-hours consumed by that job, adjusted in some cases by a multiplier associated with a special use queue:

SUs billed (node-hrs) = ( # nodes ) x ( job duration in wall clock hours ) x ( queue multiplier )

The system tracks and charges for usage to a granularity of a few seconds of wall clock time. The system charges only for the resources you actually use, not those you request. In general, your queue wait time will be less if you request only the time you need: the scheduler will have an easier time finding a slot for the 2 hours you really need than for the 48 hours you request in your job script.

Slurm Job Scheduler

Stampede2's job scheduler is the Slurm Workload Manager. Slurm commands enable you to submit, manage, monitor, and control your jobs.

Slurm Partitions (Queues)

Currently available queues include those in Stampede2 Production Queues. See KNL Compute Nodes, Memory Modes , and Cluster Modes for more information on memory-cluster modes.

Table 5. Stampede2 Production Queues

Queue Node Type Max Nodes,
(assoc'd cores per job*)
Max Duration Max jobs in queue* Charge
(per node-hour)
Configuration
(memory-cluster mode)***
development KNL 8 nodes
(544 cores)*
2 hrs 1* 1 Service Unit (SU) cache-quadrant
normal KNL 256 nodes
(17,048 cores)*
48 hrs 50* 1 SU cache-quadrant
large** KNL 2048 nodes
(139,264 cores)*
48 hrs 5* 1 SU cache-quadrant
flat-quadrant KNL 32 nodes
(2,176 cores)*
48 hrs 4* 1 SU flat-quadrant
flat-snc4 KNL 12 nodes
(816 cores)*
48 hrs 1* 1 SU flat-SNC4

* Queue status as of 16 Aug 2017. Queues and limits are subject to change without notice. Execute "qlimits" on Stampede2 for real-time information regarding limits on available queues. See Monitoring Jobs and Queues for additional information.

** To request more nodes than are available in the normal queue, submit a consulting (help desk) ticket through the TACC or XSEDE user portal. Include in your request reasonable evidence of your readiness to run under the conditions you're requesting. In most cases this should include strong or weak scaling results summarizing experiments you've run on KNL.

*** For non-hybrid memory-cluster modes or other special requirements, submit a ticket through the TACC or XSEDE user portal.

Submitting Batch Jobs with sbatch

Use Slurm's "sbatch" command to submit a batch job to one of the Stampede2 queues:

login1$ sbatch myjobscript

Here "myjobscript" is the name of a text file containing #SBATCH directives and shell commands that describe the particulars of the job you are submitting. The details of your job script's contents depend on the type of job you intend to run.

In your job script you (1) use #SBATCH directives to request computing resources (e.g. 10 nodes for 2 hrs); and then (2) use shell commands to specify what work you're going to do once your job begins. There are many possibilities: you might elect to launch a single application, or you might want to accomplish several steps in a workflow. You may even choose to launch more than one application at the same time. The details will vary, and there are many possibilities. But your own job script will probably include at least one launch line that is a variation of one of the examples described here.

These scripts are also available on Stampede2 in /share/doc/slurm. You can write your job script by copying the sample script that is most like your intended job, then editing it to meet your needs. If your job uses a software package provided by a Stampede2 module, you should also check that module's help text for additional information that may help you construct your job script.

Your job will run in the environment it inherits at submission time; this environment includes the modules you have loaded and the current working directory. You can of course use your job submission script to modify this environment by defining new environment variables; changing the values of existing environment variables; loading or unloading modules; changing directory; or specifying relative or absolute paths to files. Do not use the Slurm "--export" option to manage your job's environment: doing so can interfere with the way the system propagates the inherited environment.

The Common sbatch Options table below describes some of the most common sbatch command options. Slurm directives begin with "#SBATCH"; most have a short form (e.g. "-N") and a long form (e.g. "--nodes"). You can pass options to sbatch using either the command line or job script; most users find that the job script is the easier approach. The first line of your job script must specify the interpreter that will parse non-Slurm commands; in most cases "#!/bin/bash" or "#!/bin/csh" is the right choice. Avoid "#!/bin/sh" (its startup behavior can lead to subtle problems on Stampede2), and do not include comments or any other characters on this first line. All #SBATCH directives must precede all shell commands. Note also that certain #SBATCH options or combinations of options are mandatory, while others are not available on Stampede2.

Table 6. Common sbatch Options

Option Argument Comments
-p queue_name Submits to queue (partition) designated by queue_name
-J job_name Job Name
-N total_nodes Required. Define the resources you need by specifying either:
(1) "-N" and "-n"; or
(2) "-N" and "--ntasks-per-node".
-n total_tasks This is total MPI tasks in this job. See "-N" above for a good way to use this option. When using this option in a non-MPI job, it is usually best to set it to the same value as "-N".
--ntasks-per-node
or
--tasks-per-node
tasks_per_node This is MPI tasks per node. See "-N" above for a good way to use this option. When using this option in a non-MPI job, it is usually best to set --ntasks-per-node to 1.
-t hh:mm:ss Required. Wall clock time for job.
--mail-user= email_address Specify the email address to use for notifications.
--mail-type= begin, end, fail, or all Specify when user notifications are to be sent (one option per line).
-o output_file Direct job standard output to output_file (without -e option error goes to this file)
-e error_file Direct job error output to error_file
-d= afterok:jobid Specifies a dependency: this run will start only after the specified job (jobid) successfully finishes
-A projectnumber Charge job to the specified project/allocation number. This option is only necessary for logins associated with multiple projects.
-a
or
--array
N/A Not available. Use the launcher module for parameter sweeps and other collections of related serial jobs.
--mem N/A Not available. If you attempt to use this option, the scheduler will not accept your job.
--export= N/A Avoid this option on Stampede2. Using it is rarely necessary and can interfere with the way the system propagates your environment.

By default, Slurm writes all console output to a file named "slurm-%j.out", where %j is the numerical job ID. To specify a different filename use the "-o" option. To save stdout (standard out) and stderr (standard error) to separate files, specify both "-o" and "-e".

Launching Applications

The primary purpose of your job script is to launch your research application. How you do so depends on several factors, especially (1) the type of application (e.g. MPI, OpenMP, serial), and (2) what you're trying to accomplish (e.g. launch a single instance, complete several steps in a workflow, run several applications simultaneously within the same job). While there are many possibilities, your own job script will probably include a launch line that is a variation of one of the examples described in this section:

Launching One Serial Application

To launch a serial application, simply call the executable. Specify the path to the executable in either the PATH environment variable or in the call to the executable itself:

mycode.exe                   # executable in a directory listed in $PATH
$WORK/apps/myprov/mycode.exe # explicit full path to executable
./mycode.exe                 # executable in current directory
./mycode.exe -m -k 6 input1  # executable with notional input options

Launching One Multi-Threaded Application

Launch a threaded application the same way. Be sure to specify the number of threads. Note that the default OpenMP thread count is 1.

export OMP_NUM_THREADS=68    # 68 total OpenMP threads (1 per KNL core)
./mycode.exe

Launching One MPI Application

To launch an MPI application, use the TACC-specific MPI launcher "ibrun", which is a Stampede2-aware replacement for generic MPI launchers like mpirun and mpiexec. In most cases the only arguments you need are the name of your executable followed by any options your executable needs. When you call ibrun without other arguments, your Slurm #SBATCH directives will determine the number of ranks (MPI tasks) and number of nodes on which your program runs.

ibrun ./mycode.exe           # use ibrun instead of mpirun or mpiexec

Launching One Hybrid (MPI+Threads) Application

To launch a hybrid application, specify the number of threads per MPI rank, then use ibrun to launch the application. In general you don't need to worry about affinity: the MPI stack will distribute MPI tasks and threads in a sensible way.

export OMP_NUM_THREADS=8    # 8 OpenMP threads per MPI rank
ibrun ./mycode.exe          # use ibrun instead of mpirun or mpiexec

More Than One Serial Application in the Same Job

TACC's "launcher" utility provides an easy way to launch more than one serial application in a single job. This is a great way to engage in a popular form of High Throughput Computing: running parameter sweeps (one serial application against many different input datasets) on several nodes simultaneously. The launcher utility will execute your specified list of independent serial commands, distributing the tasks evenly, pinning them to specific cores, and scheduling them to keep cores busy. Execute "module load launcher" followed by "module help launcher" for more information.

MPI Applications One at a Time

To run one MPI application after another (or any sequence of commands one at a time), simply list them in your job script in the order in which you'd like them to execute. When one application/command completes, the next one will begin.

module load git
module list
./preprocess.sh
ibrun ./mycode.exe input1    # runs after preprocess.sh completes
ibrun ./mycode.exe input2    # runs after previous MPI app completes

More than One MPI Application Running Concurrently

(This capability is pending on Stampede2; we need to resolve some affinity issues before this approach is robust enough for production use.) To run more than one MPI application simultaneously in the same job, use ampersands to launch each instance in the background, and use the ibrun "-n" and "-o" switches to specify task counts and hostlist offsets respectively. If, for example, you use #SBATCH directives to request N=4 nodes and n=128 total MPI tasks, you will generate a hostfile with 128 entries (32 entries for each of 4 nodes). The "-n" and "-o" switches, which must be used together, determine which hostfile entries ibrun uses to launch a given application; execute "ibrun --help" for more information. Don't forget the ampersands ("&") to launch the jobs in the background, and the "wait" command to pause the script until both background tasks complete:

ibrun -n 64 -o  0 ./mycode.exe input1 &   # 64 tasks; offset by  0 entries in hostlist.
ibrun -n 64 -o 64 ./mycode.exe input2 &   # 64 tasks; offset by 64 entries in hostlist.
wait                                      # Required; else script will exit immediately.

More than One OpenMP Application Running Concurrently

You can also run more than one OpenMP application simultaneously on a single node, but you will need to distribute and pin tasks appropriately. In the example below, "numactl -C" specifies virtual CPUs (hardware threads). According to the numbering scheme for KNL hardware threads, CPU (hardware thread) numbers 0-67 are spread across the 68 cores, 1 thread per core. See TACC training materials for more information.

export OMP_NUM_THREADS=2
numactl -C 0-1 ./mycode.exe inputfile1 &  # HW threads (hence cores) 0-1. Note ampersand.
numactl -C 2-3 ./mycode.exe inputfile2 &  # HW threads (hence cores) 2-3. Note ampersand.

wait

Interactive Sessions with idev and srun

TACC's own idev utility is the best way to begin an interactive session on one or more compute nodes. To launch a thirty-minute session on a single node in the development queue, simply execute:

login1$ idev

You'll then see output that includes the following excerpts:

...
-----------------------------------------------------------------
      Welcome to the Stampede2 Supercomputer          
-----------------------------------------------------------------
...

-> After your idev job begins to run, a command prompt will appear,
-> and you can begin your interactive development session. 
-> We will report the job status every 4 seconds: (PD=pending, R=running).

->job status:  PD
->job status:  PD
...
c449-001$

The "job status" messages indicate that your interactive session is waiting in the queue. When your session begins, you'll see a command prompt on a compute node (in this case, the node with hostname c449-001). If this is the first time you launch idev, the prompts may invite you to choose a default project and a default number of tasks per node for future idev sessions.

For command line options and other information, execute "idev --help". It's easy to tailor your submission request (e.g. shorter or longer duration) using Slurm-like syntax:

login1$ idev -p normal -N 2 -n 8 -m 150 # normal queue, 2 nodes, 8 total tasks, 150 minutes

For more information see the idev documentation.

You can also launch an interactive session with Slurm's srun command, though there's no clear reason to prefer srun to idev. A typical launch line would look like this:

login1$ srun --pty -N 2 -n 8 -t 2:30:00 -p normal /bin/bash -l # same conditions as above

Interactive Sessions using ssh

If you have a batch job or interactive session running on a compute node, you "own the node": you can connect via ssh to open a new interactive session on that node. This is an especially convenient way to monitor your applications' progress. One particularly helpful example: login to a compute node that you own, execute "top", then press the "1" key to see a display that allows you to monitor thread ("CPU") and memory use.

There are many ways to determine the nodes on which you are running a job, including feedback messages following your sbatch submission, the compute node command prompt in an idev session, and the squeue or showq utilities. The sequence of identifying your compute node then connecting to it would look like this:

login1$ squeue -u bjones
 JOBID       PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
858811     development idv46796   bjones  R       0:39      1 c448-004
1ogin1$ ssh c448-004
...
C448-004$

Slurm Environment Variables

Be sure to distinguish between internal Slurm replacement symbols (e.g. "%j" described above) and Linux environment variables defined by Slurm (e.g. SLURM_JOBID). Execute "env | grep SLURM" from within your job script to see the full list of Slurm environment variables and their values. You can use Slurm replacement symbols like "%j" only to construct a Slurm filename pattern; they are not meaningful to your Linux shell. Conversely, you can use Slurm environment variables in the shell portion of your job script but not in an #SBATCH directive. For example, the following directive will not work the way you might think:

#SBATCH -o myMPI.o${SLURM_JOB_ID}   # incorrect

Instead, use the following directive:

#SBATCH -o myMPI.o%j     # "%j" expands to your job's numerical job ID

Similarly, you cannot use paths like $WORK or $SCRATCH in an #SBATCH directive.

For more information on this and other matters related to Slurm job submission, see the Slurm online documentation; the man pages for both Slurm itself ("man slurm") and its individual command (e.g. "man sbatch"); as well as numerous other online resources.

Monitoring Jobs and Queues

Several commands are available to help you plan and track your job submissions as well as check the status of the Slurm queues.

When interpreting queue and job status, remember that Stampede2 doesn't operate on a first-come-first-served basis. Instead, the sophisticated, tunable algorithms built into Slurm attempt to keep the system busy, while scheduling jobs in a way that is as fair as possible to everyone. At times this means leaving nodes idle ("draining the queue") to make room for a large job that would otherwise never run. It also means considering each user's "fair share", scheduling jobs so that those who haven't run jobs recently may have a slightly higher priority than those who have.

Monitoring Queue Status with sinfo and qlimits

To display resource limits for the Stampede2 queues, execute "qlimits". The result is real-time data; the corresponding information in this document's table of Stampede2 queues may lag behind the actual configuration that the qlimits utility displays.

Slurm's "sinfo" command allows you to monitor the status of the queues. If you execute sinfo without arguments, you'll see a list of every node in the system together with its status. To skip the node list and produce a tight summary of the available queues and their status, execute:

login1$ sinfo -o "%18P %8a %16F"    # compact summary of queue status

An excerpt from this command's output looks like this:

PARTITION          AVAIL    NODES(A/I/O/T)
development*       up       41/70/1/112
normal             up       3685/8/3/3696

The AVAIL column displays the overall status of each queue (up or down), while the column labeled "NODES(A/I/O/T)" shows the number of nodes in each of several states ("Allocated", "Idle", "Other", and "Total"). Execute "man sinfo" for more information. Use caution when reading the generic documentation, however: some available fields (e.g. TIMELIMIT, displayed using the "%l" option) are not meaningful or are misleading on Stampede2.

Monitoring Job Status with squeue

Slurm's squeue command allows you to monitor jobs in the queues, whether pending (waiting) or currently running:

login1$ squeue             # show all jobs in all queues
login1$ squeue -u bjones   # show all jobs owned by bjones
login1$ man squeue         # more info

An excerpt from the default output looks like this:

 JOBID   PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
170361      normal   spec12   bjones PD       0:00     32 (Resources)
170356      normal    mal2d slindsey PD       0:00     30 (Priority)
170204      normal   rr2-a2 tg123456 PD       0:00      1 (Dependency)
170250 development idv59074  aturing  R      29:30      1 c455-044
169669      normal  04-99a1  aturing CG    2:47:47      1 c425-003

The column labeled "ST" displays each job's status:

  • "PD" means "Pending" (waiting);
  • "R" means "Running";
  • "CG" means "Completing" (cleaning up after exiting the job script).

Pending jobs appear in order of decreasing priority. The last column includes a nodelist for running/completing jobs, or a reason for pending jobs. If you submit a job before a scheduled system maintenance or other large reservation, and the amount of time you request exceeds the time remaining until the maintenance/reservation begins, squeue will report "ReqNodeNotAvailable" ("Required Node Not Available"). The job will remain in the PD state until Stampede2 returns to production.

The default format for squeue now reports total nodes associated with a job rather than cores, tasks, or hardware threads. One reason for this change is clarity: the operating system sees each node's 272 hardware threads as "processors", and output based on that information can be ambiguous or otherwise difficult to interpret.

The default format lists all nodes assigned to displayed jobs; this can make the output difficult to read. A handy variation that suppresses the nodelist is:

login1$ squeue -o "%.18i %.9P %.8j %.8u %.2t %.10M %.6D"  # suppress nodelist

The "--start" option displays job start times, including very rough estimates for the expected start times of some pending jobs that are relatively high in the queue:

login1$ squeue --start -j 167635     # display estimated start time for job 167635

Monitoring Job Status with showq

TACC's "showq" utility mimics a tool that originated in the PBS project, and serves as a popular alternative to the Slurm "squeue" command:

login1$ showq            # show all jobs; default format
login1$ showq -u         # show your own jobs
login1$ showq -U bjones  # show jobs associated with user bjones
login1$ showq -h         # more info

The output groups jobs in four categories: ACTIVE, WAITING, BLOCKED, and COMPLETING/ERRORED. A BLOCKED job is one that cannot yet run due to temporary circumstances (e.g. a pending maintenance or other large reservation.).

If your waiting job cannot complete before a maintenance/reservation begins, showq will display its state as "WaitNod" ("Waiting for Nodes"). The job will remain in this state until Stampede2 returns to production.

The default format for showq now reports total nodes associated with a job rather than cores, tasks, or hardware threads. One reason for this change is clarity: the operating system sees each node's 272 hardware threads as "processors", and output based on that information can be ambiguous or otherwise difficult to interpret.

To cancel a pending or running job, first determine its jobid, then use scancel:

login1$ squeue -u bjones    # one way to determine jobid
   JOBID   PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
  170361      normal   spec12   bjones PD       0:00     32 (Resources)
login1$ scancel 170361      # cancel job

For detailed information about the configuration of a specific job, use scontrol:

login1$ scontrol show job=170361

To view some accounting data associated with your own jobs, use sacct:

login1$ sacct --starttime 2017-08-01  # show jobs that started on or after this date

Dependent Jobs using sbatch

You can use sbatch to help manage workflows that involve multiple steps: the "--dependency" option allows you to launch jobs that depend on the completion (or successful completion) of another job: e.g. workflows that require you to (1) compile on a single node; then (2) compute on 40 nodes; then finally (3) post-process your results using 4 nodes. This example submits a job that will run when job 173210 completes successfully:

login1$ sbatch --dependency=afterok:173210 myjobscript

For more information see the Slurm online documentation. Note that you can use $SLURM_JOBID from one job to find the jobid you'll need to construct the sbatch launch line for a subsequent one. But also remember that you can't use sbatch to submit a job from a compute node.

Visualization and Virtual Network Computing (VNC) Sessions

Stampede2 uses the KNL processors for all visualization and rendering operations. We use the Intel OpenSWR library to render graphics with OpenGL. On Stampede2, "swr" replaces "vglrun" (e.g. "swr glxgears") and uses similar syntax. OpenSWR can be loaded by executing "module load swr". We expect most users will notice little difference in visualization experience on KNL. MCDRAM may improve visualization performance for some users.

There is currently no separate visualization queue on Stampede2. All visualization apps are (or will be soon) available on all nodes. VNC sessions are available on any queue, either through the command line or via the TACC Visualization Portal. We are in the process of porting visualization application builds to Stampede2. If you are interested in an application that is not yet available, please submit a help desk ticket through the TACC or XSEDE User Portal.

Remote Desktop Access

Remote desktop access to Stampede2 is formed through a VNC connection to one or more visualization nodes. Users must first connect to a Stampede2 login node (see System Access) and submit a special interactive batch job that:

  • allocates a set of Stampede2 visualization nodes
  • starts a vncserver process on the first allocated node
  • sets up a tunnel through the login node to the vncserver access port

Once the vncserver process is running on the visualization node and a tunnel through the login node is created, an output message identifies the access port for connecting a VNC viewer. A VNC viewer application is run on the user's remote system and presents the desktop to the user.

Note: If this is your first time connecting to Stampede2, you must run vncpasswd to create a password for your VNC servers. This should NOT be your login password! This mechanism only deters unauthorized connections; it is not fully secure, as only the first eight characters of the password are saved. All VNC connections are tunnelled through SSH for extra security, as described below.

Follow the steps below to start an interactive session.

  1. Start a Remote Desktop

    TACC has provided a VNC job script (/share/doc/slurm/job.vnc) that requests one node in the vis queue for four hours, creating a VNC session.

    login1$ sbatch /share/doc/slurm/job.vnc

    You may modify or overwrite script defaults with sbatch command-line options:

    • "-t hours:minutes:seconds" modify the job runtime
    • "-A projectnumber" specify the project/allocation to be charged
    • "-N nodes" specify number of nodes needed
    • "-p partition" specify an alternate queue.

    See more sbatch options in the Common sbatch Options

    All arguments after the job script name are sent to the vncserver command. For example, to set the desktop resolution to 1440x900, use:

    login1$ sbatch /share/doc/slurm/job.vnc -geometry 1440x900

    The "vnc.job" script starts a vncserver process and writes to the output file, "vncserver.out" in the job submission directory, with the connect port for the vncviewer. Watch for the "To connect via VNC client" message at the end of the output file, or watch the output stream in a separate window with the commands:

    login1$ touch vncserver.out ; tail -f vncserver.out

    The lightweight window manager, xfce, is the default VNC desktop and is recommended for remote performance. Gnome is available; to use gnome, open the "~/.vnc/xstartup" file (created after your first VNC session) and replace "startxfce4" with "gnome-session". Note that gnome may lag over slow internet connections.

  2. Create an SSH Tunnel to Stampede2

    TACC requires users to create an SSH tunnel from the local system to the Stampede2 login node to assure that the connection is secure. On a Unix or Linux system, execute the following command once the port has been opened on the Stampede2 login node:
    localhost$ ssh -f -N -L xxxx:stampede.tacc.utexas.edu:yyyy
          username@stampede.tacc.utexas.edu

    where

    • "yyyy" is the port number given by the vncserver batch job
    • "xxxx" is a port on the remote system. Generally, the port number specified on the Stampede2 login node, yyyy, is a good choice to use on your local system as well
    • "-f" instructs SSH to only forward ports, not to execute a remote command
    • "-N" puts the ssh command into the background after connecting
    • "-L" forwards the port

    On Windows systems find the menu in the Windows SSH client where tunnels can be specified, and enter the local and remote ports as required, then ssh to Stampede2.

  3. Connecting vncviewer

    Once the SSH tunnel has been established, use a VNC client to connect to the local port you created, which will then be tunneled to your VNC server on Stampede2. Connect to localhost:xxxx, where xxxx is the local port you used for your tunnel. In the examples above, we would connect the VNC client to localhost::xxxx. (Some VNC clients accept localhost:xxxx).

    We recommend the TigerVNC VNC Client, a platform independent client/server application.

    Once the desktop has been established, two initial xterm windows are presented (which may be overlapping). One, which is white-on-black, manages the lifetime of the VNC server process. Killing this window (typically by typing "exit" or "ctrl-D" at the prompt) will cause the vncserver to terminate and the original batch job to end. Because of this, we recommend that this window not be used for other purposes; it is just too easy to accidentally kill it and terminate the session.

    The other xterm window is black-on-white, and can be used to start both serial programs running on the node hosting the vncserver process, or parallel jobs running across the set of cores associated with the original batch job. Additional xterm windows can be created using the window-manager left-button menu.

Running Applications on the VNC Desktop

From an interactive desktop, applications can be run from icons or from xterm command prompts. Two special cases arise: running parallel applications, and running applications that use OpenGL.

Running Parallel Applications from the Desktop

Parallel applications are run on the desktop using the same ibrun wrapper described above (see Running). The command:

c442-001$ ibrun ibrunoptions application applicationoptions

will run application on the associated nodes, as modified by the ibrun options.

Running OpenGL/X Applications On The Desktop

Stampede2 uses the OpenSWR OpenGL library to perform efficient rendering. At present, the compute nodes on Stampede2 do not support native X instances. All windowing environments should use a VNC desktop launched via the job script in /share/doc/slurm/job.vnc or using the TACC Vis portal.

swr: To access the accelerated OpenSWR OpenGL library, it is necessary to use the swr module to point to the swr OpenGL implementation and configure the number of threads to allocate to rendering.

c442-001$ module load swr
c442-001$ swr options application application-args

Parallel VisIt on Stampede2

VisIt was compiled under the Intel compiler and the mvapich2 and MPI stacks.

After connecting to a VNC server on Stampede2, as described above, load the VisIt module at the beginning of your interactive session before launching the Visit application:

c442-001$ module load visit
c442-001$ swr visit

VisIt first loads a dataset and presents a dialog allowing for selecting either a serial or parallel engine. Select the parallel engine. Note that this dialog will also present options for the number of processes to start and the number of nodes to use; these options are actually ignored in favor of the options specified when the VNC server job was started.

Preparing data for Parallel Visit

In order to take advantage of parallel processing, VisIt input data must be partitioned and distributed across the cooperating processes. This requires that the input data be explicitly partitioned into independent subsets at the time it is input to VisIt. VisIt supports SILO data, which incorporates a parallel, partitioned representation. Otherwise, VisIt supports a metadata file (with a .visit extension) that lists multiple data files of any supported format that are to be associated into a single logical dataset. In addition, VisIt supports a "brick of values" format, also using the .visit metadata file, which enables single files containing data defined on rectilinear grids to be partitioned and imported in parallel. Note that VisIt does not support VTK parallel XML formats (.pvti, .pvtu, .pvtr, .pvtp, and .pvts). For more information on importing data into VisIt, see Getting Data Into VisIt though this documentation refers to VisIt version 2.0, it appears to be the most current available.

Parallel ParaView on Stampede2

After connecting to a VNC server on Stampede2, as described above, do the following:

  1. Set the $NO_HOSTSORT environment variable to 1

    csh shell login1$ setenv NO_HOSTSORT 1
    bash shell login1$ export NO_HOSTSORT=1
  2. Set up your environment with the necessary modules:

    If the user is intending to use the Python interface to Paraview via any of the following methods:

    • the Python scripting tool available through the ParaView GUI
    • pvpython
    • loading the paraview.simple module into python

    then load the python, qt and paraview modules in this order:

    c442-001$ module load python qt5 paraview

    else just load the qt and paraview modules in this order:

    c442-001$ module load qt5 paraview

    Note that the qt module is always required and must be loaded prior to the paraview module.

  3. Launch ParaView:

    c442-001$ swr paraview [paraview client options]
  4. Click the "Connect" button, or select File -> Connect
  5. If this is the first time you've used ParaView in parallel (or failed to save your connection configuration in your prior runs):

    1. Select "Add Server"
    2. Enter a "Name" e.g. "ibrun"
    3. Click "Configure"
    4. For "Startup Type" in the configuration dialog, select "Command" and enter the command:

      c442-001$ ibrun swr pvserver [paraview server options]
    5. and click "Save"
    6. Select the name of your server configuration, and click "Connect"

    You'll see the parallel servers being spawned and the connection established in the ParaView Output Messages window.

Programming and Performance: General

Programming for performance is a broad and rich topic. While there are no shortcuts, there are certainly some basic principles that are worth considering any time you write or modify code.

Timing and Profiling

Measure performance and experiment with both compiler and runtime options. This will help you gain insight into issues and opportunities, as well as recognize the performance impact of code changes and temporary system conditions.

Measuring performance can be as simple as prepending the shell keyword "time" or the command "perf stat" to your launch line. Both are simple to use and require no code changes. Typical calls look like this:

perf stat ./a.out    # report basic performance stats for a.out
time ./a.out         # report the time required to execute a.out
time ibrun ./a.out   # time an MPI code
ibrun time ./a.out   # crude timings for each MPI task (no rank info)

As your needs evolve you can add timing intrinsics to your source code to time specific loops or other sections of code. There are many such intrinsics available; some popular choices include gettimeofday, MPI_Wtime and omp_get_wtime. The resolution and overhead associated with each of these timers is on the order of a microsecond.

It can be helpful to compare results with different compiler and runtime options: e.g. with and without vectorization, threading, or Lustre striping. You may also want to learn to use profiling tools like Intel VTune Amplifier ("module load vtune") or GNU gprof.

Data Locality

Appreciate the high cost (performance penalty) of moving data from one node to another, from disk to RAM, and even from RAM to cache. Write your code to keep data as close to the computation as possible: e.g. in RAM when needed, on the node that needs it. This means keeping in mind the capacity and characteristics of each level of the memory hierarchy when designing your code and planning your simulations. A simple KNL-specific example illustrates the point: all things being equal, there's a good chance you'll see better performance when you keep your data in the KNL's fast MCDRAM instead of the slower DDR4.

When possible, best practice also calls for so-called "stride 1 access" - looping through large, contiguous blocks of data, touching items that are adjacent in memory as the loop proceeds. The goal here is to use "nearby" data that is already in cache rather than going back to main memory (a cache miss) in every loop iteration.

To achieve stride 1 access you need to understand how your program stores its data. Here C and C++ are different than (in fact the opposite of) Fortran. C and C++ are row-major: they store 2d arrays a row at a time, so elements a[3][4] and a[3][5] are adjacent in memory. Fortran, on the other hand, is column-major: it stores a column at a time, so elements a(4,3) and a(5,3) are adjacent in memory. Loops that achieve stride 1 access in the two languages look like this:

Fortran exampleC example
real*8 :: a(m,n), b(m,n), c(m,n)
 ...
! inner loop strides through col i
do i=1,n
  do j=1,m
    a(j,i)=b(j,i)+c(j,i)
  end do
end do
double a[m][n], b[m][n], c[m][n];
 ...
// inner loop strides through row i
for (i=0;i<m;i++){
  for (j=0;j<n;j++){
    a[i][j]=b[i][j]+c[i][j];
  }
}

Vectorization

Give the compiler a chance to produce efficient, vectorized code. The compiler can do this best when your inner loops are simple (e.g. no complex logic and a straightforward matrix update like the ones in the examples above), long (many iterations), and avoid complex data structures (e.g. objects). See Intel's note on Programming Guidelines for Vectorization for a nice summary of the factors that affect the compiler's ability to vectorize loops.

It's often worthwhile to generate optimization and vectorization reports when using the Intel compiler. This will allow you to see exactly what the compiler did and did not do with each loop, together with reasons why.

Learning More

The literature on optimization is vast. Some places to begin a systematic study of optimization on Intel processors include: Intel's Modern Code resources; the Intel Optimization Reference Manual; and TACC training materials.

Architecture

KNL cores are grouped in pairs; each pair of cores occupies a tile. Since there are 68 cores on each Stampede2 KNL node, each node has 34 active tiles. These 34 active tiles are connected by a two-dimensional mesh interconnect. Each KNL has 2 DDR memory controllers on opposite sides of the chip, each with 3 channels. There are 8 controllers for the fast, on-package MCDRAM, two in each quadrant.

Each core has its own local L1 cache (32KB, data, 32KB instruction) and two 512-bit vector units. Both vector units can execute AVX512 instructions, but only one can execute legacy vector instructions (SSE, AVX, and AVX2). Therefore, to use both vector units, you must compile with -xMIC-AVX512.

Each core can run up to 4 hardware threads. The two cores on a tile share a 1MB L2 cache. Different cluster modes specify the L2 cache coherence mechanism at the node level.

Memory Modes

The processor's memory mode determines whether the fast MCDRAM operates as RAM, as direct-mapped L3 cache, or as a mixture of the two. The output of commands like "top", "free", and "ps -v" reflect the consequences of memory mode. Such commands will show the amount of RAM available to the operating system, not the hardware (DDR + MCDRAM) installed.

KNL Memory Modes

Figure 4. KNL Memory Modes
  • Cache Mode. In this mode, the fast MCDRAM is configured as an L3 cache. The operating system transparently uses the MCDRAM to move data from main memory. In this mode, the user has access to 96GB of RAM, all of it traditional DDR4. Most Stampede2 queues are configured in cache mode.

  • Flat Mode. In this mode, DDR4 and MCDRAM act as two distinct Non-Uniform Memory Access (NUMA) nodes. It is therefore possible to specify the type of memory (DDR4 or MCDRAM) when allocating memory. In this mode, the user has access to 112GB of RAM: 96GB of traditional DDR and 16GB of fast MCDRAM. By default, memory allocations occur only in DDR4. To use MCDRAM in flat mode, use the numactl utility or the memkind library; see Managing Memory for more information. If you do not modify the default behavior you will have access only to the slower DDR4.

  • Hybrid Mode (not available on Stampede2). In this mode, the MCDRAM is configured so that a portion acts as L3 cache and the rest as RAM (a second NUMA node supplementing DDR4).

Cluster Modes

The KNL's core-level L1 and tile-level L2 caches can reduce the time it takes for a core to access the data it needs. To share memory safely, however, there must be mechanisms in place to ensure cache coherency. Cache coherency means that all cores have a consistent view of the data: if data value x changes on a given core, there must be no risk of other cores using outdated values of x. This, of course, is essential on any multi-core chip, but it is especially difficult to achieve on manycore processors.

The details for KNL are proprietary, but the key idea is this: each tile tracks an assigned range of memory addresses. It does so on behalf of all cores on the chip, maintaining a data structure (tag directory) that tells it which cores are using data from its assigned addresses. Coherence requires both tile-to-tile and tile-to-memory communication. Cores that read or modify data must communicate with the tiles that manage the memory associated with that data. Similarly, when cores need data from main memory, the tile(s) that manage the associated addresses will communicate with the memory controllers on behalf of those cores.

The KNL can do this in several ways, each of which is called a cluster mode. Each cluster mode, specified in the BIOS as a boot-time option, represents a tradeoff between simplicity and control. There are three major cluster modes with a few minor variations:

  • All-to-All. This is the most flexible and most general mode, intended to work on all possible hardware and memory configurations of the KNL. But this mode also may have higher latencies than other cluster modes because the processor does not attempt to optimize coherency-related communication paths.

  • Quadrant (variation: hemisphere). This is Intel's recommended default, and the cluster mode in most Stampede2 queues. This mode attempts to localize communication without requiring explicit memory management by the programmer/user. It does this by grouping tiles into four logical/virtual (not physical) quadrants, then requiring each tile to manage MCDRAM addresses only in its own quadrant (and DDR addresses in its own half of the chip). This reduces the average number of "hops" that tile-to-memory requests require compared to all-to-all mode, which can reduce latency and congestion on the mesh.

  • Sub-NUMA 4 (variation: Sub-NUMA 2). This mode, abbreviated SNC-4, divides the chip into four NUMA nodes so that it acts like a four-socket processor. SNC-4 aims to optimize coherency-related on-chip communication by confining this communication to a single NUMA node when it is possible to do so. To achieve any performance benefit, this requires explicit manual memory management by the programmer/user (in particular, allocating memory within the NUMA node that will use that memory). See Managing Memory below for more information.

KNL Cluster Modes

Figure 5. KNL Cluster Modes

TACC's early experience with the KNL suggests that there is little reason to deviate from Intel's recommended default memory and cluster modes. Cache-quadrant tends to be a good choice for almost all workflows; it offers a nice compromise between performance and ease of use for the applications we have tested. Flat-quadrant is the most promising alternative and sometimes offers moderately better performance, especially when memory requirements per node are less than 16GB. We have not yet observed significant performance differences across cluster modes, and our current recommendation is that configurations other than cache-quadrant and flat-quadrant are worth considering only for very specialized needs. For more information see Managing Memory and Best Known Practices.

Managing Memory

By design, any application can run in any memory and cluster mode, and applications always have access to all available RAM. Moreover, regardless of memory and cluster modes, there are no code changes or other manual interventions required to run your application safely. However, there are times when explicit manual memory management is worth considering to improve performance. The Linux numactl (pronounced "NUMA Control") utility allows you to specify at runtime where your code should allocate memory.

When running in flat-quadrant mode, launch your code with simple numactl settings to specify whether memory allocations occur in DDR or MCDRAM. Other settings (e.g. membind=4,5,6,7) specify fast memory within NUMA nodes when in Flat-SNC-4. See TACC Training Materials for additional information.

numactl       --membind=0    ./a.out    # launch a.out (non-MPI); use DDR (default)
ibrun numactl --membind=0    ./a.out    # launch a.out (MPI-based); use DDR (default)

numactl       --membind=1    ./a.out    # use only MCDRAM
numactl       --preferred=1  ./a.out    # (RECOMMENDED) MCDRAM if possible; else DDR
numactl       --hardware                # show numactl settings
numactl       --help                    # list available numactl options

Examples. Controlling memory in flat-quadrant mode: numactl options

Intel's new memkind library adds the ability to manage memory in source code with a special memory allocator for C code and a corresponding attribute for Fortran. This makes possible a level of control over memory allocation down to the level of the individual data element. As this library matures it will likely become an important tool for those who need fine-grained control of memory.

When you're running in flat mode, the tacc_affinity script, rewritten for Stampede2, simplifies memory management by calling numactl "under the hood" to make plausible NUMA (Non-Uniform Memory Access) policy choices. For MPI and hybrid applications, the script attempts to ensure that each MPI process uses MCDRAM efficiently. To launch your MPI code with tacc_affinity, simply place "tacc_affinity" immediately after "ibrun":

    ibrun tacc_affinity a.out

Note that tacc_affinity is safe to use even when it will have no effect (e.g. cache-quadrant mode). Not also that tacc_affinity and numactl cannot be used together.

Best Known Practices and Preliminary Observations

It may not be a good idea to use all 272 hardware threads simultaneously, and it's certainly not the first thing you should try. In most cases it's best to specify no more than 64-68 MPI tasks or independent processes per node, and 1-2 threads/core. One exception is worth noting: when calling threaded MKL from a serial code, it's safe to set OMP_NUM_THREADS or MKL_NUM_THREADS to 272. This is because MKL will choose an appropriate thread count less than or equal to the value you specify. See Controlling Threading in MKL for more information. In any case remember that the default value of OMP_NUM_THREADS is 1.

When measuring KNL performance against traditional processors, compare node-to-node rather than core-to-core. KNL cores run at lower frequencies than traditional multicore processors. Thus, for a fixed number of MPI tasks and threads, a given simulation may run 2-3x slower on KNL than the same submission on Stampede1's Sandy Bridge nodes. A well-designed parallel application, however, should be able to run more tasks and/or threads on a KNL node than is possible on Sandy Bridge. If so, it may exhibit better performance per KNL node than it does on Sandy Bridge.

General Expectations. From a pure hardware perspective, a single Stampede2 KNL node could outperform Stampede1's dual socket Sandy Bridge nodes by as much as 6x; this is true for both memory bandwidth-bound and compute-bound codes. This assumes the code is running out of (fast) MCDRAM on nodes configured in flat mode (450 GB/s bandwidth vs 75 GB/s on Sandy Bridge) or using cache-contained workloads on nodes configured in cache mode (memory footprint < 16GB). It also assumes perfect scalability and no latency issues. In practice we have observed application improvements between 1.3x and 5x for several HPC workloads typically run in TACC systems. Codes with poor vectorization or scalability could see much smaller improvements. In terms of network performance, the Omni-Path network provides 100 Gbits per second peak bandwidth, with point-to-point exchange performance measured at over 11 GBytes per second for a single task pair across nodes. Latency values will be higher than those for the Sandy Bridge FDR Infiniband network: on the order of 2-4 microseconds for exchanges across nodes.

MCDRAM in Flat-Quadrant Mode. Unless you have specialized needs, we recommend using tacc_affinity or launching your application with "numactl --preferred=1" when running in flat-quadrant mode (see Managing Memory above). If you mistakenly use "--membind=1", only the 16GB of fast MCDRAM will be available. If you mistakenly use "--membind=0", you will not be able to access fast MCDRAM at all.

Affinity. Default affinity settings are usually sensible and often optimal for both threaded codes and MPI-threaded hybrid applications. See TACC training materials for more information.

MPI Initialization. Our preliminary scaling tests with Intel MPI on Stampede2 suggest that the time required to complete MPI initialization scales quadratically with the number of MPI tasks (lower case "-n" in your Slurm submission script) and linearly with the number of nodes (upper case "-N").

Programming and Performance: Phase 2 System (SKX)

Pending. This material will be available during the Phase 2 SKX early user period.

File Operations

Under Construction. Until this material is available here, consult the Stampede1 User Guide

Revision History

"Last Update" at the top of this document is the date of the most recent change to this document. This revision history is a list of non-trivial updates; it excludes routine items such as corrected typos and minor format changes.

Click to view