Fwd: This has been brought up by CRC at our last renewal and delayed it...please make every effort to acknowledge them in any work we crank out. Our ability to continue using CRC may depend on this
Artur Dubrawski
awd at cs.cmu.edu
Fri Jul 28 10:25:22 EDT 2023
THis is an important note for everyone working on some of our healthcare
projects with Pitt which utilize their CRC data hosting and computing
resources.
We are expected to acknowledge in our papers and presentations these two
additional sources of funding - see the specific NIH and NSF award info
below.
Cheers
Artur
---------- Forwarded message ---------
From: Clermont, Gilles <cler at pitt.edu>
Date: Fri, Jul 28, 2023 at 8:48 AM
Subject: This has been brought up by CRC at our last renewal and delayed
it...please make every effort to acknowledge them in any work we crank out.
Our ability to continue using CRC may depend on this
To: Hauskrecht, Milos <milos at pitt.edu>, Parker, Bob <rparker at pitt.edu>,
Kyle Miller <mille856 at andrew.cmu.edu>, awd at cs.cmu.edu <awd at cs.cmu.edu>,
Pinsky, Michael R <pinsky at pitt.edu>, Al-Zaiti, Salah Shafiq <SSA33 at pitt.edu>
*From:* Center for Research Computing <bgc14 at crc.pitt.edu>
*Sent:* Friday, July 28, 2023 8:03 AM
*To:* Clermont, Gilles <cler at pitt.edu>
*Subject:* CRC News July 2023
Acknowledge CRC resources, MPI error alerts, new job efficiency display
Is this email not displaying correctly? View it in your browser
<https://centerforresearchcomputing.cmail20.com/t/j-e-ekuthkk-iuuhvhyjk-r/>
[image: University of Pittsburgh]
*Pitt Research *
*Center for Research Computing *
CRC News July 2023
------------------------------
Researchers required to acknowledge NIH S10 and NSF MRI awards when using
CRC resources
CRC’s recent hardware upgrades are thanks to two very large grants from NIH
and NSF. Proper citation plays a crucial part in the University being able
to secure more grants of this nature in the future. For all publications
that use research conducted at the CRC after July 2022 please follow the
instructions below (also found on the CRC website:
https://crc.pitt.edu/about/acknowledge
<https://centerforresearchcomputing.cmail20.com/t/j-l-ekuthkk-iuuhvhyjk-j/>
).
All research products (publications, conference proceedings, and
presentations) resulting from an allocation of computing time or storage on
HTC (HTC cluster) should include the following acknowledgement:
*"This research was supported in part by the University of Pittsburgh
Center for Research Computing, RRID:SCR_022735, through the resources
provided. Specifically, this work used the HTC cluster, which is supported
by NIH award number S10OD028483."*
All research products (publications, conference proceedings, and
presentations) resulting from an allocation of computing time or storage on
H2P (SMP, MPI, and GPU clusters) should include an acknowledgement:
*"This research was supported in part by the University of Pittsburgh
Center for Research Computing, RRID:SCR_022735, through the resources
provided. Specifically, this work used the H2P cluster, which is supported
by NSF award number OAC-2117681."*
You are free to also acknowledge or thank individual members of the CRC
team.
Please remember to also list any research products (publications,
conference proceedings, and presentations) resulting from usage of CRC
resources in any allocation or renewal proposal.
------------------------------
Some MPI Cluster Users Will See Minimum Node Count Error
Users of the MPI cluster may notice the following output when attempting to
submit a job requesting less than 2 nodes:
*sbatch: error: ERROR: Minimum node count is 2!*
* sbatch: error: Batch job submission failed: Node count specification
invalid*
This is to prevent single-node jobs that should otherwise be run on the SMP
cluster from using MPI-centric hardware that were configured specifically
to support parallel jobs across multiple compute nodes.
To prevent similar undersubscription at the CPU level and for consistency
with accounting of service unit usage for MPI jobs, a warning message is
currently in place for jobs that request CPU cores less than the full
amount that the node offers (ntasks-per-node):
sbatch: WARNING: Your MPI job is not requesting the full number of cores on
its nodes: 48 cores on the mpi partition. MPI jobs will need to request the
full number of cores to be successfully submitted to SLURM after July 15,
2023. Please modify your submission script to request the full core amount
or consider submitting your job to the SMP cluster.
After July 15, MPI jobs that do not request the full node's resources will
no longer enter the queue. You will need to adjust your SLURM submission
script to request all of the cores available. At a minimum, your job
submission script will need the following SLURM directives:
*#SBATCH --nodes=2*
* #SBATCH --ntasks-per-node=48 #SBATCH --cluster=mpi #SBATCH
--partition=mpi*
If your MPI workflow is optimized for a certain number of cores, you will
need to make sure that the proper amount is being specified to your
software by input argument, configuration file parameter, etc. Please submit
a help ticket
<https://centerforresearchcomputing.cmail20.com/t/j-l-ekuthkk-iuuhvhyjk-t/>
if you need guidance in updating your job submission script.
------------------------------
Ask Nick: Can I access information on my job’s efficiency?
After CRC made a policy change on the lower limit of requested CPUs on MPI
(see details above), some users were still submitting and being allocated
multiple full nodes worth of CPUs, but only a small number were being
utilized for the duration of the job. We used a script provided in the
SLURM installation called "seff" to display this for an individual job, and
I created a wrapper around that functionality to show output more easily
for multiple jobs.
The idea is to equip users with information about just how much of their
job's resources are being used over the duration of the job, given their
submission script configuration. This output could be used by PIs to
demonstrate the efficiency of the submitted jobs when they request a yearly
allocation of service units. This functionality will be available in
mid-August.
Users will be able to specify which cluster, which user, and a date range
to look for completed jobs. The most valuable information is in the CPU
Efficiency and Memory Efficiency items. When these are high, the full
allocation of resources for the job is being used. Low values indicate that
the user is requesting more resources than necessary for their processing,
and that they should adjust their submission script. The function currently
only shows CPU output, but we hope to add efficiency for GPU usage in the
future.
In the example below of output for multiple jobs, the top line is the input
command, followed by what the user sees regarding their completed jobs.
Note that one job failed, and therefore the efficiency could not be
calculated.
*crc-effective -M mpi -u leb140 -S 06/01/23*
* Showing jobs on mpi that completed between 06/01/23 and now for leb140:*
*cluster name: mpi*
* Job ID: 1489891 Cluster: mpi User/Group: leb140/sam State: COMPLETED
(exit code 0) Nodes: 2 Cores per node: 48 CPU Utilized: 2-08:55:19 CPU
Efficiency: 99.10% of 2-09:26:24 core-walltime Job Wall-clock time:
00:35:54 Memory Utilized: 4.45 MB Memory Efficiency: 0.00% of 187.50 GB*
*cluster name: mpi*
* Job ID: 1490283 Cluster: mpi User/Group: leb140/sam State: FAILED (exit
code 1) Cores: 1 CPU Utilized: 00:00:00 CPU Efficiency: 0.00% of 00:00:00
core-walltime Job Wall-clock time: 00:00:00 Memory Utilized: 0.00 MB
(estimated maximum) Memory Efficiency: 0.00% of 1.95 GB (1.95 GB/core)*
*Nickolas Comeau is a CRC Research Computing Specialist*
------------------------------
[image: University of Pittsburgh]
<https://centerforresearchcomputing.cmail20.com/t/j-l-ekuthkk-iuuhvhyjk-h/>
University of Pittsburgh
Center for Research Computing
312 Schenley Place
4420 Bayard Street
Pittsburgh, PA 15260
412-648-3094
crc.pitt.edu
<https://centerforresearchcomputing.cmail20.com/t/j-l-ekuthkk-iuuhvhyjk-k/>
You are receiving this email as a user of the Center for Research Computing
Unsubscribe
<https://centerforresearchcomputing.cmail20.com/t/j-u-ekuthkk-iuuhvhyjk-y/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/autonlab-users/attachments/20230728/d39e5405/attachment-0001.html>
More information about the Autonlab-users
mailing list