There are several changes to CCR's systems for the spring semester.  These are detailed below.  If you have any questions or problems, please submit a ticket to CCR help.


1.  Change to default wall time on academic cluster as of 1/21/2020:

The DEFAULT wall time on the UB-HPC (academic) cluster is changing from 72 hours to 24 hours.  Users may still request the maximum wall time of 72 hours.  However, if unspecified in your job script or OnDemand interactive desktop session, Slurm will assume you want the default of only 24 hours.  This can be modified in the job script using:

#SBATCH --time=01:00:00  (format is hours:minutes:seconds)


THIS IS REALLY IMPORTANT!  If you don't specify the wall time in your job script, you won't get an error.  However, your job will only run for 24 hours, even if it hasn't completed in that time.  Analysis of the jobs submitted to the ub-hpc cluster shows a large majority of those using the default wall time of 72 hours actually ran for less than 24 hours!  The job scheduler works best for everyone when users specify only the time they need for their jobs to run.  By shrinking the default wall time on the cluster, the scheduler will run more efficiently.  Certainly, if you need more than 24 hours, you can specify up to the maximum wall time of 72 hours.




2. New addition to the academic cluster - available 1/29: 

NOTE: these are not currently available in OnDemand interactive apps.  Users must submit batch jobs to access them

 

We are happy to announce the addition of new compute nodes acquired as part of a $1 million grant from the National Institutes of Health.  We are adding 128 new compute nodes to the UB-HPC (academic) cluster including general computational nodes, larger memory nodes, and GPU nodes. All nodes are contained in one partition – cascade – in the UB-HPC cluster. Below are the specifications for the nodes:
 

---------------

Partition: cascade

---------------

total nodes: 128

Slurm tags:  UBHPC,CPU-Gold-6230,INTEL,NIH,IB

 

General nodes:

total nodes:  96

gpu_count:  0

cores_per_node:  40

memory size: 187 GB

product name:  Dell PowerEdge C6420

SLURM CPU constraint:  --constraint=CPU-Gold-6230

SLURM memory setting:  --mem=187000

 

GPU nodes:

total nodes:  8

gpu_count:  2 (V100)

cores_per_node:  40

memory size:  187 GB

product name:  Dell PowerEdge R740

SLURM CPU constraint:  --constraint=CPU-Gold-6230

SLURM memory setting:  --mem=187000

Additional Slurm tag:  V100

 

Large memory nodes:

total nodes:  24

gpu_count:  0

cores_per_node:  40

memory size:  754 GB

product name:  Dell PowerEdge C6420

SLURM CPU constraint:  --constraint=CPU-Gold-6230

SLURM memory setting:  --mem=754000

 

3. Priority boosts for PIs listed on the NSF MRI (2018) and NIH (2019) grants


Faculty members that are listed on the grants that enabled us to purchase the skylake and cascade lake nodes now have access to priority boosts for their group.  PIs with an active allocation to the UB-HPC cluster in ColdFront, have been given access to QOS values that provide a priority boost to their group's jobs that run on those nodes.  You can see what QOS (quality of service) boost you've been given using the 'slimits' command when logged into the cluster.  These also show up as allocations in ColdFront.  More details can be found HERE.  If you do not have an allocation for the QOS boost but believe you should, please contact CCR Help.  The 'nih' and 'nihsupport' qos values are valid for use on the 'cascade' partition and the 'mri' and 'mrisupport' qos values are valid for use on the skylake, largemem, and gpu partitions.  More details on these priority boosts and how to use a QOS can be found HERE.


Are you a PI on the grant but don't have a CCR account? - follow these instructions to get one!


PIs can not automatically renew priority boost allocations.  Please see this article for more info



4.  Any researcher with active funding from NIH is eligible for the NIH priority boost

PIs MUST enter grant information in ColdFront and have an active allocation to the UB-HPC cluster.  After verification of the funding, CCR will provide access to the QOS for the priority boost.  Instructions for adding grant information to ColdFront can be found HERE.  Please contact CCR help after this is complete as providing this access is a manual process.  



5.  Please acknowledge resources provided by CCR!  Wording and citation info can be found HERE


 

6.  CCR help desk expands


We're happy to announce the addition of three undergraduate student help desk assistants this semester!  They will be answering tickets, updating documentation, and helping expand our help desk services.  Currently, they are located with the majority of CCR staff at the downtown campus but we're hoping to open a walk-up help desk on north campus in the future.   Phone support information will be forthcoming.