techstaff:slurm
Differences
This shows you the differences between two versions of the page.
Both sides previous revisionPrevious revisionNext revision | Previous revisionNext revisionBoth sides next revision | ||
techstaff:slurm [2016/01/05 09:01] – [More] kauffman | techstaff:slurm [2020/10/07 13:08] – [More] kauffman | ||
---|---|---|---|
Line 1: | Line 1: | ||
- | ====== | + | ====== Peanut Job Submission Cluster ====== |
- | We are currently **alpha** testing and gauging user interest in a cluster of machines that allows for the submission of long running compute jobs. Think of these machines as a dumping ground for discrete computing tasks that might have been rude or disruptive to execute on the main (shared) shell servers (i.e., linux1, linux2, linux3). | + | Think of these machines as a dumping ground for discrete computing tasks that might be rude or disruptive to execute on the main (shared) shell servers (i.e., linux1, linux2, linux3). |
- | For job submission we will be using a piece of software called [[http:// | + | For job submission we will be using a piece of software called [[http:// |
"Slurm is an open-source workload manager designed for Linux clusters of all sizes. It provides three key functions. First it allocates exclusive and/or non-exclusive access to resources (computer nodes) to users for some duration of time so they can perform work. Second, it provides a framework for starting, executing, and monitoring work (typically a parallel job) on a set of allocated nodes. Finally, it arbitrates contention for resources by managing a queue of pending work." | "Slurm is an open-source workload manager designed for Linux clusters of all sizes. It provides three key functions. First it allocates exclusive and/or non-exclusive access to resources (computer nodes) to users for some duration of time so they can perform work. Second, it provides a framework for starting, executing, and monitoring work (typically a parallel job) on a set of allocated nodes. Finally, it arbitrates contention for resources by managing a queue of pending work." | ||
- | SLURM is similar to most other queue systems in that you write a batch script, then submit it to the queue manager. The queue manager schedules your job to run on the queue (or partition in SLURM parlance) that you designate. Below is an outline of how to submit jobs to SLURM, how SLURM decides when to schedule your job, and how to monitor progress. | + | Slurm is similar to most other queue systems in that you write a batch script, then submit it to the queue manager. The queue manager schedules your job to run on the queue (or partition in Slurm parlance) that you designate. Below is an outline of how to submit jobs to Slurm, how Slurm decides when to schedule your job, and how to monitor progress. |
+ | |||
===== Where to begin ===== | ===== Where to begin ===== | ||
- | SLURM is a set of command line utilities that can be accessed via the command line from **most** any computer science system you can login to. Using our main shell servers (linux.cs.uchicago.edu) is expected to be our most common use case, so you should start there. | + | Slurm is a set of command line utilities that can be accessed via the command line from **most** any computer science system you can login to. Using our main shell servers (linux.cs.uchicago.edu) is expected to be our most common use case, so you should start there. |
ssh user@linux.cs.uchicago.edu | ssh user@linux.cs.uchicago.edu | ||
+ | ===== Mailing List ===== | ||
+ | If you are going to be a user of this cluster please sign up for the mailing list. Downtime and other relevant information will be announced here. | ||
+ | |||
+ | [[ https:// | ||
===== Documentation ===== | ===== Documentation ===== | ||
- | The [[http:// | + | The [[http:// |
- | A great way to get details on SLURM commands are the manuals that are already on the cluster. For example, if you type the following command: | + | A great way to get details on Slurm commands are the manuals that are already on the cluster. For example, if you type the following command: |
man sbatch | man sbatch | ||
you will get the manual page for the '' | you will get the manual page for the '' | ||
===== Resources ===== | ===== Resources ===== | ||
- | * [[https:// | + | * [[https:// |
- | * [[http:// | + | * [[http:// |
- | * [[http:// | + | * [[http:// |
- | * [[http:// | + | * [[http:// |
* [[https:// | * [[https:// | ||
+ | * [[http:// | ||
Line 34: | Line 41: | ||
==== Hardware ==== | ==== Hardware ==== | ||
Our cluster contains nodes with the following specs: | Our cluster contains nodes with the following specs: | ||
+ | |||
+ | '' | ||
* 16 Cores (2x 8core 3.1GHz Processors), | * 16 Cores (2x 8core 3.1GHz Processors), | ||
* 64gb RAM | * 64gb RAM | ||
* 2x 500GB SATA 7200RPM in RAID1 | * 2x 500GB SATA 7200RPM in RAID1 | ||
- | To better manage the cluster we have virtualized the job submission nodes and give them all resources of the hardware. So, the actual resources you can consume on any one node is: | + | '' |
- | * 14 Cores, | + | * 24 Cores (2x 24core Intel Xeon Silver 4116 CPU @ 2.10GHz), 48 threads |
- | * 62GB RAM | + | * 128gb RAM |
+ | * OS: 2x 240GB Intel SSD in RAID1 | ||
+ | * /local: 2x 960GB Intel SSD RAID0 | ||
+ | '' | ||
+ | * 6 Cores (Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz), 12 threads | ||
+ | * 32gb RAM | ||
+ | * OS: 1x 512gb SSD | ||
+ | |||
+ | '' | ||
+ | * 16 Cores (Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz), 32 threads | ||
+ | * 128gb RAM | ||
+ | * OS: 2x Samsung SSD 850 PRO 128GB | ||
+ | * /local: ZFS mirror (2x Samsung SSD 850 PRO 1TB) | ||
+ | * 2x Quadro P4000 | ||
+ | |||
+ | '' | ||
+ | * 8 Core (Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz), 16 threads | ||
+ | * 64gb RAM | ||
+ | * OS: 1x 1TB 7200k spinning disk. | ||
+ | * 4x GeForce GTX 1080 Ti | ||
==== Storage ==== | ==== Storage ==== | ||
There is slow scratch space mounted to '' | There is slow scratch space mounted to '' | ||
Line 58: | Line 86: | ||
Request interactive shell | Request interactive shell | ||
< | < | ||
+ | |||
+ | Create a directory on the scratch partition if you don't already have one: | ||
+ | < | ||
Change into my scratch directory: | Change into my scratch directory: | ||
- | < | + | < |
Get the files I need: | Get the files I need: | ||
< | < | ||
- | user@research2:/ | + | user@slurm1:/ |
foo | foo | ||
</ | </ | ||
Check that the file now exists: | Check that the file now exists: | ||
< | < | ||
- | user@research2:/ | + | user@slurm1:/ |
-rw------- 1 user user 105121 Dec 29 2015 foo | -rw------- 1 user user 105121 Dec 29 2015 foo | ||
</ | </ | ||
Line 77: | Line 108: | ||
== Performance is slow == | == Performance is slow == | ||
This is expected. The maximum speed this server will ever be able to achieve is 1Gb/s because of its single 1G ethernet uplink. If this cluster gains in popularity we plan on upgrading the network and storage server. | This is expected. The maximum speed this server will ever be able to achieve is 1Gb/s because of its single 1G ethernet uplink. If this cluster gains in popularity we plan on upgrading the network and storage server. | ||
+ | |||
==== Utilization Dashboard ==== | ==== Utilization Dashboard ==== | ||
Sometimes it is useful to see how much of the cluster is utilized. You can do that via the following URL: http:// | Sometimes it is useful to see how much of the cluster is utilized. You can do that via the following URL: http:// | ||
Line 88: | Line 120: | ||
| **debug** | The partition your job will be submitted to if none is specified. The purpose of this partition is to make sure your code is running as it should before submitting a long running job to the general queue. | | | **debug** | The partition your job will be submitted to if none is specified. The purpose of this partition is to make sure your code is running as it should before submitting a long running job to the general queue. | | ||
| **general** | All jobs that have been thoroughly tested can be submitted here. This partition will have access to more nodes and will process most of the jobs. If you need to use the '' | | **general** | All jobs that have been thoroughly tested can be submitted here. This partition will have access to more nodes and will process most of the jobs. If you need to use the '' | ||
+ | | **fast** | 2019-12-02: 48 threads, 128GB RAM | | ||
+ | | **quadro** | 2019-12-02: 2x Quadro P4000. *| | ||
+ | | **pascal** | 2018-05-04: 1x Nvidia GTX1080.| | ||
+ | | **titan** | 2018-05-04: 4x Nvidia GTX1080Ti. *| | ||
+ | * This partition is shared and you MUST use the '' | ||
====== Job Submission ====== | ====== Job Submission ====== | ||
Jobs submitted to the cluster are run from the command line. Almost anything that you can run via the command line on any of our machines in our labs can be run on our job submission server agents. | Jobs submitted to the cluster are run from the command line. Almost anything that you can run via the command line on any of our machines in our labs can be run on our job submission server agents. | ||
Line 99: | Line 135: | ||
===== Command Summary ===== | ===== Command Summary ===== | ||
[[http:// | [[http:// | ||
- | | ^ SLURM ^ Example ^ | + | | ^ Slurm ^ Example ^ |
^ Submit a batch serial job | sbatch | sbatch runscript.sh | | ^ Submit a batch serial job | sbatch | sbatch runscript.sh | | ||
- | ^ Run a script | + | ^ Run a script |
^ Kill a job | scancel | scancel 4585 | | ^ Kill a job | scancel | scancel 4585 | | ||
^ View status of queues | squeue | squeue -u cnetid | | ^ View status of queues | squeue | squeue -u cnetid | | ||
Line 108: | Line 144: | ||
===== Usage ===== | ===== Usage ===== | ||
- | Below are some common examples. You should consult the [[http:// | + | Below are some common examples. You should consult the [[http:// |
+ | |||
+ | === Default Quotas === | ||
+ | By default we set a job to be run on one CPU and allocate 100MB of RAM. If you require more than that you should specify what you need. Using the following options will do: '' | ||
=== Exclusive access to a node === | === Exclusive access to a node === | ||
Line 114: | Line 153: | ||
==== sbatch ==== | ==== sbatch ==== | ||
- | The sbatch command is used for submitting jobs to the cluster. sbatch accepts a number of options either from the command line, or (more typically) from a batch script. An example of a SLURM batch script is shown below: | + | The sbatch command is used for submitting jobs to the cluster. sbatch accepts a number of options either from the command line, or (more typically) from a batch script. An example of a Slurm batch script is shown below: |
=== Sample script === | === Sample script === | ||
Make sure you create a directory in which to deposit the '' | Make sure you create a directory in which to deposit the '' | ||
- | mkdir -p $HOME/ | + | mkdir -p $HOME/ |
< | < | ||
#!/bin/bash | #!/bin/bash | ||
# | # | ||
- | #SBATCH --mail-user=user@cs.uchicago.edu | + | #SBATCH --mail-user=cnetid@cs.uchicago.edu |
#SBATCH --mail-type=ALL | #SBATCH --mail-type=ALL | ||
- | #SBATCH --output=/ | + | #SBATCH --output=/ |
- | #SBATCH --error=/ | + | #SBATCH --error=/ |
- | #SBATCH --workdir=/ | + | #SBATCH --workdir=/ |
#SBATCH --partition=debug | #SBATCH --partition=debug | ||
#SBATCH --job-name=check_hostname_of_node | #SBATCH --job-name=check_hostname_of_node | ||
#SBATCH --nodes=1 | #SBATCH --nodes=1 | ||
#SBATCH --ntasks=1 | #SBATCH --ntasks=1 | ||
+ | #SBATCH --mem-per-cpu=500 | ||
#SBATCH --time=15: | #SBATCH --time=15: | ||
Line 140: | Line 180: | ||
man sbatch | man sbatch | ||
- | Make sure to replace all instances of the word '' | + | Make sure to replace all instances of the word '' |
+ | === Submitting job script === | ||
+ | Using the above example you will want to place your tested code into a file. ' | ||
+ | < | ||
+ | sbatch hostname.job | ||
+ | </ | ||
+ | |||
+ | You can then check the status via squeue or see the output in the output directory ' | ||
==== srun ==== | ==== srun ==== | ||
Used to submit a job to the cluster that doesn' | Used to submit a job to the cluster that doesn' | ||
< | < | ||
user@host: | user@host: | ||
- | research2 | + | slurm2 |
- | research2 | + | slurm2 |
</ | </ | ||
Line 172: | Line 219: | ||
==== sinfo ==== | ==== sinfo ==== | ||
- | View information about SLURM nodes and partitions. | + | View information about Slurm nodes and partitions. |
The following code block shows the what happens when you run the '' | The following code block shows the what happens when you run the '' | ||
Line 178: | Line 225: | ||
user@host: | user@host: | ||
PARTITION AVAIL TIMELIMIT | PARTITION AVAIL TIMELIMIT | ||
- | debug* | + | debug* |
- | general | + | fast up 1-00: |
+ | general | ||
+ | pascal | ||
+ | quadro | ||
+ | titan up 3-00: | ||
</ | </ | ||
Line 185: | Line 236: | ||
====== Monitoring Jobs ====== | ====== Monitoring Jobs ====== | ||
- | '' | + | '' |
Running '' | Running '' | ||
Line 203: | Line 254: | ||
This command: | This command: | ||
- | srun -p general --pty --mem 500 -t 0-06:00 /bin/bash | + | srun -p general --pty --cpus-per-task 1 --mem 500 -t 0-06:00 /bin/bash |
will start a command line shell ('' | will start a command line shell ('' | ||
+ | |||
====== Job Scheduling ====== | ====== Job Scheduling ====== | ||
+ | We use a [[http:// | ||
- | We use a [[http:// | + | We also have backfill turned on. This allows for jobs which are smaller to sneak in while a larger higher priority job is waiting for nodes to free up. If your job can run in the amount of time it takes for the other job to get all the nodes it needs, |
- | + | ||
- | We also have backfill turned on. This allows for jobs which are smaller to sneak in while a larger higher priority job is waiting for nodes to free up. If your job can run in the amount of time it takes for the other job to get all the nodes it needs, | + | |
Line 218: | Line 269: | ||
| JOB < | | JOB < | ||
| Job < | | Job < | ||
- | | JOB < | + | | JOB < |
- | | error: Unable to allocate resources: More processors requested than permitted | It usually has **nothing** to do with priviledges | + | | error: Unable to allocate resources: More processors requested than permitted | It usually has **nothing** to do with privileges |
+ | |||
+ | ====== Using the GPU ====== | ||
+ | |||
+ | ===== GRES Multiple GPU's on one system ===== | ||
+ | GRES: Generic Resource. As of 2018-05-04 these only include GPU' | ||
+ | |||
+ | Jobs will not be allocated any generic resources unless specifically requested at job submit time using the '' | ||
+ | < | ||
+ | |||
+ | Jobs will be allocated specific generic resources as needed to satisfy the request. If the job is suspended, those resources do not become available for use by other jobs. | ||
+ | |||
+ | Job steps can be allocated generic resources from those allocated to the job using the '' | ||
+ | |||
+ | ==== Ok, but I don't want to read the wall of text above ==== | ||
+ | Fine. | ||
+ | |||
+ | The '' | ||
+ | |||
+ | < | ||
+ | --gpu=gpu: | ||
+ | # Please try to limit yourself to one GPU per person. | ||
+ | </ | ||
+ | |||
+ | Example when using tensorflow: | ||
+ | |||
+ | Given the file '' | ||
+ | < | ||
+ | # | ||
+ | from tensorflow.python.client import device_lib | ||
+ | print(device_lib.list_local_devices()) | ||
+ | </ | ||
+ | |||
+ | Here we can see that no GPU was allocated to us because we did not specify the '' | ||
+ | < | ||
+ | user@bulldozer: | ||
+ | user@gpu3: | ||
+ | user@gpu3: | ||
+ | </ | ||
+ | |||
+ | If we request only 1 GPU. | ||
+ | < | ||
+ | user@bulldozer: | ||
+ | user@gpu3: | ||
+ | physical_device_desc: | ||
+ | </ | ||
+ | |||
+ | If we request 2 GPUs. | ||
+ | < | ||
+ | user@bulldozer: | ||
+ | user@gpu3: | ||
+ | physical_device_desc: | ||
+ | physical_device_desc: | ||
+ | </ | ||
+ | |||
+ | If we request more GPUs then are available. | ||
+ | < | ||
+ | kauffman3@bulldozer: | ||
+ | srun: error: Unable to allocate resources: Requested node configuration is not available | ||
+ | </ | ||
+ | |||
+ | ==== Cool, but how do I know where and what resources are available ==== | ||
+ | Turns out the '' | ||
+ | < | ||
+ | $ sinfo -O partition, | ||
+ | PARTITION | ||
+ | debug* | ||
+ | fast slurm[9-14] | ||
+ | general | ||
+ | pascal | ||
+ | quadro | ||
+ | titan | ||
+ | </ | ||
+ | |||
+ | FEATURES: Is actually just an arbitrary string in the configuration file that defines a node. However, techstaff hopes it actually provides some useful info. | ||
+ | |||
+ | GRES: Don't depend on this being accurate, however it will definitely give you a clue as to how many generic resources are in a partition. | ||
+ | |||
+ | |||
+ | ==== Checking how many Generic RESources are being consumed ==== | ||
+ | |||
+ | Simple use the '' | ||
+ | < | ||
+ | $ squeue -O username, | ||
+ | USER NODELIST | ||
+ | someusername | ||
+ | otherusername | ||
+ | ... | ||
+ | </ | ||
+ | |||
+ | |||
+ | ===== Environment Variables ===== | ||
+ | |||
+ | ==== CUDA_HOME, LD_LIBRARY_PATH ==== | ||
+ | |||
+ | Please make sure you specify $CUDA_HOME and if you want to take advantage of CUDNN libraries you will need to append / | ||
+ | |||
+ | cuda_version=9.2 | ||
+ | export CUDA_HOME=/ | ||
+ | export LD_LIBRARY_PATH=$LD_LIBRARY_PATH: | ||
+ | |||
+ | Currently we support the same versions of CUDA that the latest version of CUDNN supports. This is not written in stone and we can accommodate most other versions if required; just let techstaff know what your needs are. | ||
+ | |||
+ | ==== PATH ==== | ||
+ | You may also need to add the following to your '' | ||
+ | |||
+ | export PATH=$PATH:/ | ||
+ | |||
+ | ==== CUDA_VISIBLE_DEVICES ==== | ||
+ | Do not set this variable. It will be set for you by Slurm. | ||
+ | |||
+ | The variable name is actually misleading; since it does NOT mean the amount of devices, but rather the physical device number assigned by the kernel (e.g. / | ||
+ | |||
+ | For example: If you requested multiple gpu's from Slurm (--gres=gpu: | ||
+ | |||
+ | |||
+ | ===== Example ===== | ||
+ | This sbatch script will get device information from the installed Tesla gpu. | ||
+ | < | ||
+ | # | ||
+ | # | ||
+ | #SBATCH --mail-user=cnetid@cs.uchicago.edu | ||
+ | #SBATCH --mail-type=ALL | ||
+ | #SBATCH --output=/ | ||
+ | #SBATCH --error=/ | ||
+ | #SBATCH --workdir=/ | ||
+ | #SBATCH --partition=gpu | ||
+ | #SBATCH --job-name=get_tesla_info | ||
+ | |||
+ | export PATH=$PATH:/ | ||
+ | export LD_LIBRARY_PATH=$LD_LIBRARY_PATH=/ | ||
+ | |||
+ | cat << EOF > / | ||
+ | #include < | ||
+ | |||
+ | int main() { | ||
+ | int nDevices; | ||
+ | |||
+ | cudaGetDeviceCount(& | ||
+ | for (int i = 0; i < nDevices; i++) { | ||
+ | cudaDeviceProp prop; | ||
+ | cudaGetDeviceProperties(& | ||
+ | printf(" | ||
+ | printf(" | ||
+ | printf(" | ||
+ | | ||
+ | printf(" | ||
+ | | ||
+ | printf(" | ||
+ | | ||
+ | } | ||
+ | } | ||
+ | EOF | ||
+ | |||
+ | / | ||
+ | / | ||
+ | rm / | ||
+ | rm / | ||
+ | </ | ||
+ | ==== Output ==== | ||
+ | STDOUT will look something like this: | ||
+ | < | ||
+ | cnetid@linux1: | ||
+ | Device Number: 0 | ||
+ | Device name: Tesla M2090 | ||
+ | Memory Clock Rate (KHz): 1848000 | ||
+ | Memory Bus Width (bits): 384 | ||
+ | Peak Memory Bandwidth (GB/s): 177.408000 | ||
+ | </ | ||
+ | STDERR should be blank. | ||
+ | ====== Feedback ====== | ||
+ | If you feel this documentation is lacking in some way please let techstaff know. Email [[techstaff@cs.uchicago.edu]], | ||
====== More ====== | ====== More ====== | ||
- | If you feel this documentation is lacking | + | Sometimes other universities have documentation |
+ | |||
+ | - [[ https:// | ||
+ | |
/var/lib/dokuwiki/data/pages/techstaff/slurm.txt · Last modified: 2021/01/06 16:13 by kauffman