Jobs reports 'excessive memory usage' message
A batch job output may contain warnings about excessive memory usage.
A batch job output may contain warnings about excessive memory usage.
We will have rolling reboots of Owens and Pitzer clusters including login and compute nodes, starting from Monday, February 3, 2020
We have found that large MPI jobs may hang at startup with mvapich2/2.3
and mvapich/2.3.1
(on any compiler dependency) due to a known bug that has been fixed in release 2.3.2. If users experience this issue, please switch to mvapich2/2.3.2
.
We are having rolling reboots of Owens cluster including login and compute nodes, starting from October 10, 2019.
NCBI blocks any connection from computing nodes because they are behind firewalls. Thus OSC users cannot use SRA tools to download data "on-the-fly" at runtime on computing nodes, e.g. 'fastq-dump -X 5 SRR390728'. OSC users must download SRA data on login using the command 'prefetch' before any sequence analysis. Please see the section 'Download SRA Data' in the SRA Toolkit software page for more detail.
We have found that recent MPI jobs using openmpi/1.10-hpcx and openmpi/2.0-hpcx on Owens may complete or hang until the job is killed, but receive segmentation fault. Some applications might be affected (if you run these applications with openmpi mentioned above): orca, openfoam and lammps. OSC users can use other compatible versions, e.g. openmpi/1.10.7-hpcx and openmpi/2.1.6-hpcx. Please check available version from the software page.
We will have rolling reboots of all three clusters starting 9:30 AM June 05, 2019.
Extra 'pmlogger' messages during job startup and ending on Owens and Pitzer on March 11, 2019
Gaussian-4 (G4) theory calculations in Gaussian 16 Rev. B.01 can produce erratic results. A workaround is to use Gaussian 16 Rev. A.03 or Rev. C.01, e.g.:
module load gaussian/g16c01
Gaussian 16 Rev. A.03 has been made the default module on all clusters. A typical symtom is energy blow up in step 8 resulting in an unusually large, erroneous HF/GFHFB2 energy and consequently an incorrect G4 total energy.
We will perform the replacement work of Ethernet switches from 12pm to 3pm on Thursday, Jan 17, which includes all login nodes and 2 quick nodes on Owens. As a result, users won't be able to log into Owens at the beginning and end of the maintenance work, and won't be able to use Owens VDI through OnDemand during the entire maintenance window. Running jobs on Owens, as well as other OSC services (Pitzer, Ruby, and fileystems) won't be impacted.