Members: David, Dawn, Dee, Ed, Hamp, Jim
J., Mark S., Mark McKeown, Michael, Robin, Stan, Sue Ellen, Tim S.,
Tim T., Tom S.,
Convener: Tim T.
In Attendance: Dee, Ed, Hamp, Jim, Mark S.,
Tang Tang (for Robin), Tim S., Tim T., Tom S.
Recorder: Tim T.
Next Meeting: July 25, 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: Aliases for teal cluster is in place, connecting to
"teal.unix" round-robins between teal1 to 3.unixlab. No similar
arrangement for orange cluster because hoping to use Sun's load
balancing software for this.
Ed: PBS 2.3 is not out yet for Solaris 2.7, still not working on
orange cluster. And it's sluggish responding on teal cluster, it
doesn't monitor frequently enough to stop someone from submitting
lots of jobs in quick succession and overloading one node. This is
suppose to be fixed in 2.3. The vendor said 3 weeks ago it'd be out
imminently.
Also, teal cluster needs a cronjob to clean out /tmp regularly
Research Computing Funds Allocation
Suggestion list e-mailed prior to meeting:
- Help Andrew Grimshaw/Legion with additional disk capacity for Centurion and well as funds for maintenance and repair.
- Funds for software for Centurions, such as compilers, extending licenses for popular research software (e.g, Matlab) to Centurions, etc.
- Help with funds to prevent half the nodes on Centurion being shutdown over the summer since no money was allocated by the university to fix the AC in Small 106 where Centurion is housed. NOTE: Hamp and I are working with Andrew on both short (move some nodes to ITC's mainframe machine room) and long term solutions to this.
- Money for training, perhaps in System Administration, Security, or something similar? I have no plan, but would be willing to help develop some possibilities.
- Additional disk capacity, including RAID array.
We dropped numbers 3 and 4 (Centurion cooling and training) as not appropriate expenditures for these funds. Other funding sources should be found for these items, though, because they are important to be done.
Discussed the Sun MindPrint proposal and their Sun Ray technology and it's appropriateness for our lab environment. The Ultra-60s do not have 3-D rendering.
Tom: crick is heavily used, may need more cycles there. ACHS is putting together
a beowolf cluster.
Hamp: SGIs are being used
Jim: Strategic change if we want to put together a Linux cluster, not sure
we need it yet. Does Ansys have Linux version?
Hamp: No Linux version of ANSYS. Also with Linux, 2.4 kernal supports 32bit
UIDs but it's not yet out. And once it comes out, then the applications have
to be revised to support them as well.
Tim S.: what if we add SMP nodes with single processor nodes to the SP? Power3
nodes?
Hamp: may not need to, next generation of nodes from IBM will be dual-chip.
But we should look at SMP nodes as well.
Discussed feasibility of dropping software from SP, such as SPSS and S-Plus and running on another platform - such as NT Server. Partially based on SPSS's announced intention to drop Unix support/upgrades. Hard to tell if this would work for our users and not sure what back-end we could put it on to support their use. Is it needed? What about BIG jobs that can't run on workstation?
Hamp: Could expand the HSM capacity, quadruple capacity for
acrchive capacity funds.
Also local storage on SP nodes with no time limit until job
completes, then when it completes a finite amount of time to get
files moved before hteyr automatically cleaned off. Bill is looking
into doing this and tying it into LoadLeveler job. We could make
better use of local storage this way.
Tim S.: How much rack (frame) space is left on SP.
Hamp: Room for six more thin nodes or 3 wide ones.Tim S.: Could you get an estimate
on the maximum computing power we could put in those 6 slots, with some of them
SMP?
Tom: crick is out of disk space
Jim/Tim S.: We can help with that.
Tom: Raid array?
Hamp: Definitely want it to be RAID
Tom: $10K-12K will get 288 GB RAID -he'll get prices on several configurations.
Tim S.: Could any of these be in by fall?
Jim: Anything we need to do on the SGI side?
Tom: Mendel is an 0-200; we do need more SGIs.
Tim S.: Check into upgrading four slowest nodes on SP.
SP
Load Limit change was enacted.
Ed: Tried to get usage statistics, but none available yet for April-June. A 200hour job was in the queue for 2 weeks. The throughput is lengthening with the load limit change.
Tim S.: The limit was only increased to 2, right?
Hamp: Yes.
HSM
Tim S.: what are our policies for exporting the HSM?
Hamp: Same as with home directory-some hosts we count as secure,
some not, and we export accordingly. BTW, there are 16 users on
home2 with special exports; not wedge users.
They are reviewing Home Directory quotas, the quota trees, user
quotas
Research Computing Support Center (RCSC)
The Center had a Brown Bag series on new SAS version 8. Had 5 users
attend; good questions, interest, enthusiasm for the event. Brown
Bags will become a regular monthly event in the fall.
New Business
Meeting adjourned at 11:40, next meeting July 25, 2000, 10:30 AM
Go to: ITC Research Computing
Standing Committee Home Page
by Tim F.J. Tolson