Minutes for the ITC-Research Computing Management Meeting
May 22, 2006 at 10:30AM
(2nd floor conference room, 2015 Ivy Rd)
Members: Alice, Hamp, Jim, Terry, Tim S., Tom S. Tim T.
Attending: Alice, Hamp, Jim, Terry, Tim S., Tom S., Tim T.
Chair: Tim T. Recorder: Alice Howard
I. Corrections to minutes from last Management meeting on April 6, 2006? None.
II. Ongoing Discussion Topics:
· “Condo” Cluster #2 –aka dogwood:
o The Dell equipment is here – but still waiting for all the pieces (e.g. racks, etc) to arrive – and the electrical wiring is coming along. It still might be possible to keep to the timeline agreed to at the December meeting: have it set up and configured in June 2006, and advertise it as “go live” in Summer 2006.
· Cedar Cluster:
o Still waiting to hear about the new version of AVAKI and their reimplementation – Steve is pinging them
· Aspen & Birch Clusters:
o Will turn off Aspen once Dogwood is up and running – and will wait to decide about exact timing.
· Oak Cluster:
o Can move to interactive, round-robin login most anytime. Will target this for mid-June and Hamp will check with Chip about preferred dates.
· Teak Cluster:
o SGI has filed for bankruptcy.
· Proposal for Linux cluster support as ‘for-fee’ service. At December meeting, Hamp said he’d translate current hourly Unix support rates into table/information showing cost for supporting cluster of X size for Y years.
o It’s hard to make these estimates, but Hamp will try to make a table so that users can do comparisons.
· Acquiring monitoring tools as discussed in March, especially MPI Link Checker (65 to 128 nodes is $4250 less 20% = $3400).
o Hamp will check to see if this has been ordered.
· Andew Grimshaw’s request for Linux front-end to campus grid project:
o Hamp has several outstanding queries out to Andrew – he’ll ping him again.
· Update on transition from clear text passwords to Eservices for Unix logins/connections to blue.unix, to HSM & Longtmp:
o This will be done for all unix systems – or none.
o Currently working on a Kerberos solution but having problems with incorporating ‘groups’ from the LDAP server.
o This transition will involve a major user education issue – especially for blue.unix users.
o No timeline for this project yet – it will wait until after the password synchronization website is up and working.
· Any news on UVa’s participation in responding to NSF’s Request for Proposals for next generation/iteration of Supercomputing Centers?
o Made it through the first round.
· Mitch Rosen’s IT – Infrastructure Supporting Research Task Force. (https://www1.seas.virginia.edu/itrtf/) (Maillist is itrtf@virginia.edu)
o Early draft is circulating for input.
o Secondary group of faculty and ITC staff working on CI-TEAM NSF proposal response. Approved Katherine and Ed 1/8 time each to work on grant if UVa gets it.
· Update on CSS Research Computing support staff and services relocating to Brown Science & Engineering Library and Alderman Library fourth floor in summer 2006:
o The move to Brown Science and Engineering Library will be in mid-late June; the move to Alderman in late July or early August.
o Will be doing joint marketing for both new centers with the Library – and also participating with the Library in Reunions weekend in June and Parents weekend in the Fall
III. New Business
· Mitch’s email about MAE faculty that wants 24-30 nodes with Infiniband:
o This could be done as a part of the Dogwood cluster – or they could keep it separate and just pay for system administration services.
· “Priority “ access for participants in condo clusters:
o Our agreement specifies “priority access to at least 80% of the cycles of your nodes” – where priority access means that your jobs will go to the front of the queue when submitted but will not preempt currently executing jobs.
o But participants sometimes want their priority to mean that their jobs will execute now.
o We need to look at this again – what do we think they should get? There needs to be a real benefit that can be described (and measured?) in order for the condo model to work.
o One way to approach this might be to use the Super Computer Center allocation-of-resources model which establishes a base allocation with “tickets” for allocations above that.
o Meanwhile, the new Dogwood cluster is big enough so that its increased capacity might give us some breathing space to look at alternatives to the “priority problem.”
Next Scheduled Meeting of whole Standing Committee group: Monday, June 26, 2006, 10:30 AM in ITC-2015 Ivy Road, Room #102.
Next Scheduled Meeting of ITC ResComp Management team: Monday, July 24, 2006, 10:30 AM in ITC-2015 Ivy Road, Room #220.
Go to: ITC Research Computing Standing Committee Home Page