Research Support
Policy and Procedures for the UMDNJ SunFire 6900 High Performance Computing Resource known as the “Scientific Server”.
The Scientific Server is a shared resource. The SunFire is equipped with 24 x 1.2 GHz dual-core ultrasparc 4 cpus. At present this resource is subdivided into 3 domains. Domain A has 4 cpus; Domain B has 4 cpus and Domain C has 16 cpus. Domain C is known as the Scientific Server (aka siriusC). The SunFire is fibre-linked to an 8 TB SAN. At present, approximately 800 GB (1 tray) of the SAN is dedicated to siriusC. SiriusC has 14 cpus (28 cores) dedicated to the grid queue for advanced computation and 2 cpus (4 cores) dedicated to EMBOSS and system administration.
Accounts Policy. The scientific server will be restricted to use by UMDNJ faculty, staff and students. Outside collaborators may also use the resource pending a review process. The following types of accounts exist:
Research Account – Provides 10 GB of quota on the SAN. Research accounts may require a formal review for approval.
Student Account – Provides 500 MB of quota on the SAN. The student account is for graduate students in GSBS taking courses where high performance software tools are required. These accounts can be upgraded to research accounts pending review.
See http://istweb.umdnj.edu/research_services/sci_server.shtml for more information about obtaining an account on the scientific server.
Batch job restrictions on the UMDNJ SunFire 6900:
The batch queue is maintained by Sun Grid Engine. See Sun Grid Engine Primer for more information (http://istweb.umdnj.edu/research_services/pdf_files/UMDNJ_SGE_Primer.pdf )
Time limit: 24 hours (the max time you can request is 24:00:00!)
cpu limit per job: There is no limit to the number of cores per job we can enforce. However, there is a physical limit of 28 total cores dedicated to the batch queue at this time. Use the qstat command to monitor usage of cpu on the queue. If you request more cores than are available, your job will hang in queue. Likewise, if you request more time than the time limit allows per job, your job will hang in queue indefinitely.
Maximum No. of Jobs: 8 (You may have no more than 8 jobs running in the queue at the same time.)
Avoid requesting more cpu/cores than the physical limit; otherwise your job may hang in queue unnecessarily.
