Members: David Drake, Dawn, Dee, Ed, Hamp, Jim J., Mark S., Michael Black, Robin, Stan, Sue Ellen, Tim S., Tim T., Tom S.,
Convener: Tim T.
In Attendance: David, Dawn, Ed, Hamp, Jim, Mark, Michael, Robin, Sue Ellen,Tim S., Tim T.
Recorder: Tim T.
Next Meeting: March 27, 2000 at 10:30 AM in Astronomy 117.

Click here to see the agenda from this meeting.

To Dos & Old Business:

Research Computing platforms and SP

Hamp: We need about $22.5K for SGI Indy replacement in UnixLab.
ToDo(Tim T): check with Terry whether this much available in lab upgrade budget

Ed- Report on console ogins in UnixLabs.
Should we re-configured some of UnixLab Suns as part of the Orange cluster for batch operation?
Make SGIs an interactive cluster.
Make Suns a batch cluster with some additional UnixLab machines.
Hamp: Realize the Ultra10's are three times faster than Ultra1's in current UnixLab
Ed: We could go to CPU limit on interactive use, and queues for batch.
Hamp- Have we talked to professors who might be teaching or using labs for classroom or instruction? Fundamental question is: Do we need Unix-based public lab?
Dawn: Jeff and I worked with professor to get software on SGIs in Thornton E225. Seems like E225 is used frequently.
Robin: Small Hall used by students in classes but none are taught there.
Ed: When UnixLabs were originally created there was no Exceed for the PCs. When E225 Thornton created, the Stacks PC lab didn't exist and no electronic classrooms.
Jim: do we have data on Exceed usage?
Tim S.: Could we replace Unix workstations with PCs?
Mark: one problem would be that some CAD design sofware doesn’t run on PCs, only Unix workstations.
Tom: If we went with PCs we may need to pay for Exceed 3D, which may be need for high end graphic work demanded.
ToDo(Tim T): Contact Mitch Rosen and Aylor on usage of UnixLabs. Suggest a survey of E-School faculty and graduate students. Also should check with LSPs Melissa and Diane.

Orange/Teal cluster: Ed: Rollout of Orange Cluster-with current 6 nodes could only have one 8 hour class
Teal cluster: 02 compute engines with no console logins?

Jim J. Do we know the current load on the Origins?
Ed: 2 jobs - 1 job running 3 months; other 2 machines not used we could rollout 2 nodes on PBS batch, need to leave at least one for interactive use such as GCG.
Hamp: 1 faculty in Astronomy would use John Hawley uses SGI elsewhere. Others use San Diago Center. Could use to develop code, and then run their jobs at San Diego; also GCG users could run their jobs.
Ed: If Teal to be all interactive, need load balancing for logons or a web page that shows current load like we have for Suns in UnixLab.

Tim S:-could spend the $100K allocated for research computing at the end of this fiscal year. Susie is still looking for ETF allocations. If next year’s budget request of $150K gets funded and Susie carries over, we'd have a combined to have $250K, a quarter-million!

Tom: Could we spend some of this on software compilers, debuggers - user tools to make it easier to use the hardware we're providing?
Tim T: We've been allocated around $8,000 from Terry for research-specific software.
Hamp: Need to also consider SUR grants, even though we didn't participate this year, the program still exists. The E-school and Commerce did get participate. So do we want to participate or try to this coming year? Or should we consider using funds to leverage Sun's Higher Education grants and similar program for SGIs?
Tim S.: What kind compute cycles do we need?
Hamp: Bill P. working on SP parallel job stats, but Hamp has gotten them yet. Hamp provided a handout on cost comparison of some possible compute engines: various model IBM, Sun, SGIs.
Jim: Where is parallel use? Is it time to put together an Intel-based cluster – Linux or Solaris?
Tim S: What about getting cycles from Grimshaw's Intel-based Linux cluster?
Tom: we talked about taking/using half of his cycles. He's currently only using about 2% of the cycles.
Ed: Some in E-school using them.
Tim S: what about faster serial jobs?
Tom: what about quicker communication among parallel nodes? The parallel nodes not getting used because you have to write MPI code.
Ed: Not always, IMSL has a parallel version. Abaqus is suppose to have a parallel version soon. More will use parallel nodes as more comonly used software come out with parallel versions.
Dawn: We also need good tools for researchers to use parallel, like a parallel code debugger.
Tim S. Sounds like 90% of the jobs are serial jobs, so most of the dollars should go to provide fastest serial computing possible.
Tom: And need to consider other needs, not just cycles, what about disk storage? HMS working?
Tom S. Are there disk bottlenecks, do we need RAID array?
Jim J. Is there other parallel software we should get?
Dawn: We're getting Totalview as the parallel debugger, is there parallel code writing software?
Jim: What is used most for this? Fortran or C++?
Dawn: Mostly Fortran
Tim S.: Should we poll users, get input? Make a list of options to present as a starter?
Tom: Users mostly use defaults, don't explore compile or other options that might speed up their code. For example, KAP pre-processor (Kuck & Associates) for Fortran or C. It unrolls loops and increases speed. We have lots of un-used tools - but how to get researchers using them. KAP pre-processor is a separately licensed package, with one floating license. Should we make something like this a default invocation for xlf?
Dawn: Need to try to inform researchers, use ITC Research News, to inform and get input.
David : could we target users, find out the users Fortran and C – and e-mail them about compiler options, tips and tricks?
Hamp: We do have some recent license logs for Sun, but no recent IBM.
ToDo(Hamp): check on usage/license logs, especially Sun.
Try to have some ideas for funds allocation/needs by next meeting:
ToDo(Tim T): Start an "idea" list e-mail.

Specifically SP
Tim S. Sounds like hardware is less important.
Jim: Need to gather usage, how heavily used are the SP nodes?
Ed: typically there are a couple of jobs waiting in the queue.
Tom: He has noticed that ACHS workstations not as loaded as in the past.
Tim S. Have we talked to the Pilkey (Auto Safety lab) group?
Ed: Tom and Chris worked with and he contacted Steve Reagan when eight nodes were configured for parallel use. Never heard back. Ed( Steve Reagan contacted-
Dawn: Did the new DYNA 3D get installed? Where is the new version?
ToDo (Dawn): She'll check on, make sure we (ITC) isn't the hold-up on getting this up and running.
ToDo (Dawn) She'll see if there's an auto-parallelizing tool.
Hamp: The new version of Amber is parallel, are there other packages, Maple, MatLab, SAS?
ToDo (Ed): He'll see if Maple, Matlab have parallel versions coming.

HSM
Dawn : Had user ask about using HMS from a SAS job.
ToDo (Dawn) Contact user to see if actually using SAS on HMS.
Hamp: Two blue.unix nodes have been upgraded to AIX 4. 3 nicknamed the "yellow cluster”. He's migrated all the maggie users before end of year except some public directories, old Ken Scully archives?
ToDo (Hamp/Dawn): Decide if need any of these files.
Hamp: The Special Collections files, he copied all he could get off, about 62 GB. Should we store these data or the Library? He's working with the Library-they're getting lots of online storage space.
Hamp: Doesn't think we should have a send off for maggie. When we rolled over to 2000, cron broke under AIX 3.2., but nothing about UniTree that was still working broke!

Software
ToDo (Tim T./Tom) Get together and decide about TotalView licenses and payment, from RCS and site license budget and for what platforms.

Discussion on Visualization software:
Tom: Only one user of Iris Explorer on their workstation, because we export the tree it's on.
Ed: Over the last year, there's been 2 users, 1 consistent user. Gary Shiflett. It is a visual programming environment
Group decided that we will drop after this year
Data Explorer: is now freeware, last time we contacted, they didn't have the latest version available for AIX. It's freeware now, so we should just let users who want it go to their website and download and we don't provide any support.
Khoros: Tom has latest version free, but there's no upgrade. The BME people use it on their machines. We should document that there's no upgrade and recommend IDL as the replacement.

New Business
None

Meeting adjourned at 12:00, next meeting March 27, 2000, 10:30 AM

Go to: ITC Research Computing Standing Committee Home Page

by Tim F.J. Tolson