Notes
for the ITC-Research Computing Standing Committee Meeting
November 10, 2008 at 10:30 AM
2015 Ivy Road, 1st Floor Conference Room
Present: Andrew, Jim, Joe S., Mark, Robin,
Tim S., Tom, Katherine Holcomb, Bill Pemberton
Chair: Tom
Recorder: Alice (Tom, for this meeting)
I. Corrections to minutes
from last meeting on October 13, 2008? (Click to review the agenda
from this meeting)
II. Old Business:
- Update on CSAC, NSF PetaScale Track 2 meetings, and related. - No updates
- Storage needs assessment -- LSP dialog, survey (Joe) – No updates
- Cluster node purchase by faculty – revisit for those using non-ETF funds (Mike) – Just before the meeting came an announcement that ETF funds have been restored for FY08-09. It still is possible that there will be no ETF funds for FY09-10. ITC will postpone budget decisions until this is known. Our recommended configuration continues to be 8 cores (4x2) with 32 GB of RAM.
- Possibility of acquired "gently used" systems – WeÕre pretty sure we donÕt want the Cray from Aberdeen.
- Communication with potential benefactor –
No updates
- Update on new machine-room building – No updates
III. New Business:
- Gig Ethernet Policy – Mike Johnson in
Jordan made a request for two gigabit ports; his LSP contacted NetOps who
aid ITC is in a spending freeze.
Even with the restoration of ETF funds, ITC will wait until the
GovernorÕs budget is submitted (around 12/20) before deciding how to
proceed with network upgrades.
Tom to contact Mike Johnson.
- Any other Items?
- 1) PBS policies. The APB is trying to address the dissatisfaction that
some condo owners are expressing about not being able to get to the Òfront
of the lineÓ. PBS can suspend
jobs, although this will require more virtual memory. If this doesnÕt work, itÕs okay to
terminate running jobs.
Whatever we do, the ITC web site needs to explain whatÕs happening
and why (and at whose direction).
There was discussion about banning all sequential jobs on the
cluster. ITC needs the APB to
communicate any of these choices in writing. It was pointed out that the cedar nodes are more than
three years old, so those owners are no longer entitled to priority in the
queue. There are zero or one
dogwood condo owners. Bill
will use a test cluster to determine whether PBS can suspend and restart
an MPI job. Is MPI smart
enough to suspend all instances?
There was some discussion of having PBS set resource limits.
- 2) The high-performance computing grid is running
in some public labs. This
might become the preferred platform for serial jobs, or lower limits might
be set for cluster jobs (again, ITC needs this in writing). To use the lab machines
effectively, the ÒkillÓ policy currently in place would need to be
changed. This in turn would
require bigger page files, a change in the reboot policy (currently early
every morning); 64-bit Windows would also be needed. Tom will talk to Stephanie Conley
about the possibilities.
IV. Adjourn by 11:45 and next meeting
- Next Scheduled Meeting of ITC ResComp Standing
Committee: December 8, 2008 in ITC-2015 Ivy Road, Room #102 (First Floor
Conference Room). Due to a
permanent scheduling conflict, the time for this meeting is being changed
to 10:45 a.m.
Please send suggestions, additions, corrections to: Tom or Alice
