| Login | |
|
Issues/Circulation
Current Issue: June 2006 Past Issues
Education
Research
MyMCSR on the Web
Introducing MyMCSR
Supercomputers
Software |
Volume 5, Number 1 Redwood Upgraded to SUSE In January 2006, MCSR upgraded redwood's O/S from Red Hat based SALE Linux to Novell's SUSE Linux. The software upgrade itself went well, and all applications tested out ok without having to be recompiled. In the months following the upgrade, a few users have reported mysterious "Address Error" messages in the output of their prematurely terminated Gaussian 03 output files on redwood. We are collecting statistics on these occurrences, so if this happens to you, please email us the job id and the absolute pathnames of your g03 input and output files, and don't erase them. We will probably recompile G03 on redwood in June, possibly in conjunction with the hardware upgrade, and then we will attempt to rerun these jobs to see if this fixes the Address Error.In conjunction with, but as a separate item from, the SUSE upgrade, SGI engineers also replaced 70 power boards on the Altix, to address some identified problems. This involved substantial disassembly and reassembly of the supercomputer. It is not uncommon for such radical maintenance to introduce new problems, and for it to take two or more iterations to get everything back into proper working order. That is in fact what happened in January. Later in the semester, unexpected reboots of redwood were performed on March 8, March 20, and March 27. When redwood is rebooted for monthly maintenance on Friday, May 5, it will have been running without a reboot for 5 1/2 weeks. There have been intermittent network interface outages to redwood over that period of time, but all have been short, and none have affected running jobs. An O/S patch will be applied to the system during the May 5 maintenance window, which should correct the interface outage problem. In other redwood news, a new queue, Big-Red-2, has been added to the PBS instance governing the slower redwood processes (the instance where Red-2 resides.) Big-Red-2 is for jobs that need only one or 2 processors but up to 12GB of memory. If you have a job that fits the bill, please email us your justification, so that we can enable your id to submit jobs to Big-Red-2. |
