Job Submission Issues Resolved

Starting in late October and into the beginning of November numerous reports of slow or hung job submission requests were reported. The scheduling engine, LSF,  was modified to perform multi-threaded job submission processing which appears to have resolve the issue. Since implementing the change we not seen a resurgence of the issue on the cluster. 

Tips and Tricks for Job Submission

To request a GPU node on interactive mode, you can use this sample command:  bsub -q gpu-v100 -W 2:00 -n 8 -M 32000 -gpu "num=1" -Is bash

The gpu-v100 queue can utilize up to 2 gpu's per job.

The best way to maximize the use of a node while launching a job, should be like this example: bsub -W <time> -n <cores> -M <memory> -R "span[hosts=1]" <program>

If you need X-Forwarding on a job, use: bsub -W <time> -n <cores> -M <memory> -XF -Is bash 

Firewall Patching Nov 11th (4-7am) 

The CCHMC security team will be updating the internal firewall appliances that separate the various networking segments within the Research bubble. This upgrade is expected to be low impact due to the firewalls redundant implementation but some precautions are suggested. For this reason we recommend users not schedule the following job functions during this time: 

1. Any workload processing or writing data hosted on an rds6 volume. 
2. Any job transferring data into or out of CCHMC. 

Please also note that any active connections to the cluster (citrix, SSH) may experience a short pause or require a reconnect during this window as the upgrade is performed. Please plan your workload accordingly. 

New R project Version

We've recently added R version 4.1.1 to the cluster.   Load it with:

$ module load R/4.1.1

Please send any package requests to help-cluster@bmi.cchmc.org.