Present: Rob Cameron, Ashley Deacon, Richard Dubois, Imre Kabai, David MacFarlane, Stuart Marshall, Randy Melen, Marc Messerschmidt, Brian Moritz, Leonard Moss, Amedeo Perazzo, Vineet Rawat, Norm Ringgold.
Via teleconference: Amber Boehnlein.

Meeting agenda page

Minutes:

  • Introduction to/from David MacFarlane
    • Discussion of SCSC's role, its connection to the Executive Council
  • Rob gave update on IT SubCouncil
    • Revamped SLACspace home with improved organization
    • Two (of four) meetings remain for cataloging IT services
      • Better online documentation required (turning spreadsheet into documentation)
  •  Upcoming hardware acquisitions:
    •  PPA:
      • Richard on Fermi:
        • $800K of hardware (all storage, no compute)
        • Lots of tape (1500 at least)
      • Richard on BaBar:
        • 24 more nodes (on top of existing 4)
        • Len correctly said it is 60 by end of fiscal year
      • Richard on ATLAS:
        • $400K, mostly storage, maybe 30 compute nodes
      • Stuart on KIPAC:
        • In need of storage
        • Pending Lance's investigation
        • 2-3 trays, 10-20U
        • 10U of compute to go into Orange(?) cluster
      • Richard on EXO and CDMS:
        • small purchases, ~30TB each 
    • Amedeo on LCLS:
      • Looking for JBOD with SAS, a few different vendors being considered
      • SSN, SanMina, AMAX
      • No new compute nodes, upgrades to processors instead, compute nodes underused
    • Brian on Photon Science:
      • Depends on grants
      • Tom: $100-$150K, about a rack
        • Colfax solution, Kerry getting Dell pricing
      • Jenz may want SunCap expansion of a rack
      • Norm indicates the troubles of planning with respect to grant cycles
    • Ashley on SSRL:
      • None
    • Vineet on AD:
      • Yunhai Cai's cluster
      • Len thinks Les Cotrell may want to add to that cluster (some code being tested on SUNCAT?)
  • Richard on recharge
    • Agreement with ALDs was for everything, power, racks, etc. but some things aren't budgeted for
      • e.g., Fermi wants 10Gbps, 1Gbps is standard, how to charge them more?
    • Small clusters
      • 4-node cluster in SSRL, $20K/year cluster charge plus per-node charge seems excessive
      • Crossover point?
      • Len: need to create incentive to contribute to large clusters
    • Norm's thoughts:
      • Dispel the rumors about capacity
        • removal of silos should give 1700-1800 additional sq. ft.
        • 1 MW residual power
        • Sufficient chilled water (without getting 10" pipes)
      • Need to plan for delivery and distribution
        • Needs lead time
        • Without sufficient lead time, it might seem like a capacity issue to whoever's making the request
      • Need Boris + 1 more, but getting only 50% of Boris (other 50% to building 50)
    • Discussion between Richard, Norm, Stuart, David about how to encourage good behavior (e.g., keeping storage sufficiently dense)
      • Economic incentives? Charge for low-density, as in the 10Gbps case?
    • David: Sandy agreed to pay considerable amount, situation is not sustainable.  Need to prepare for unforseen costs.
  • Richard on ALD retreat
    • 1.5h session on sci. comp.
    • Discussion of different "end-to-end" sci. comp. models in the various directorates
      • Slide seems to fit HEP model
      • Can Brian replicate the slide for simulation group
    • David: any assistance for proposal writing (SciDAC, etc.)? Campus advertises such help
      • Richard to check with Amber
    • Discussion of two governance models:
      • collaborative
        • agreement that this is more appropriate for sci. comp.
        • Richard needs to articulate convincing defense
        • Imre: perhaps show how the SubCouncil approach would break down
      • reporting to one
  • Seminar series in need of name so that advertising can begin (posters to be printed)

Actions:

  • Richard to clean up slides for ALD retreat
  • Brian to provide simulation group's workflow for ALD retreat slide
  • Imre to discuss ALD retreat slides further with Richard
  • SCSC members to come up with potential seminar series names

Post-meeting email updates:

Thomas Eriksson on sci-comp mailing list:

I received a big list of initial members from Anders Borglund, and I have collected and handful of people at SSRL that would be interested in the sci-comp mailing list.

Please see if some of you at AD, PS, and LCLS could gather some initial members, and we could go ahead and create the mailing list.

Vineet on SC seminar series:

  • Seminars will be held monthly.
    • We'll go off-schedule occasionally to accommodate out-of-town speakers who might be visiting.
    • Time and location does not have to be strictly fixed (at this frequency people might forget or not care).
  • A number of great potential speakers have been proposed from both inside and outside SLAC.
    • Initially, talks closely related to SLAC research are preferred.
    • General sci. comp. related external talks are important to the series' success.  Maybe about a third of all speakers should be external?
      • Possible speakers from LBL, eBay, Facebook, Wikimedia, UCs Berkeley and Davis, Pixar, DESY, ...
  • The proposed timeline for speakers is as follows:
    • End of April (or as soon as Amber is here): Garth Williams (LCLS Coherent X-ray Imaging) - confirmed
    • Late May: Bill Dally (TBD?) (NVIDIA Chief Scientist & Sr. VP of Research, Stanford CS Prof.)
    • Mid-June: Ralf Kaehler (Cosmology visualization), possibly including a tour of Kavli's 3D vis. room) - confirmed
    • Late June (coordinated with the SC Jamboree): UC Davis Professor on Memristors
    • End of June: Mike Huffer (LSST storage), possibly Anton Barty (LCLS Imaging) who's interested but may not be available in early June
  • Kavli auditorium is preferred location.
    • Limited availability, we need to set dates and book it soon.
  • Still searching for an appropriate time (maybe we need a doodle poll?):
    • Monday ruled out because of weekly colloquium at 4PM.
    • Tuesday might work.
    • Wednesday ruled out because of weekly LCLS seminar (around 3PM); general dislike of the 1:30PM slot at the Mar. 28 SCSC meeting.
    • Thursday at 3PM seemed to work, but conflicts were reported at the Mar. 28 SCSC meeting.
    • Friday ruled out because of proximity to the weekend.
  • Seminar series also needs a name (posters need to be printed soon).
    • Homework for SCSC members.

Imre on FudForum status:

FudForum has been installed, fs. backups and db. exports completed. Crowd integration is working and Tony is looking into email integration.

Tony sent out a detailed list of suggestions around governance, I will add more to his email and try to bring together the right people to talk about the governance aspects. I have Tony, Farnaz, Randy, Richard, Ernest and Teresa in mind. Please let me know, if you want to be involved and/or know someone who should be involved.

  • No labels