To Dos & Old Business:
Numbers in parenthesis match the numbers on the To-Do Table of March 29, 1999 that we reviewed at this meeting. Item 1: Research Computing Engines ---------------------------------- Hamp: Concerned that current names already exist and are known by a few users (o200-#.sp2, (#=1,2,3) and changing to teal# might confuse users. We decided the name change needed to be done and that it was a way to announce the transition to loadbalancing and queue management software. Hamp: SUN does not have separate loadbalancing software, so that's out as a possibility. Their "loadleveling" software is bundled with their "true cluster" machines, the U450s and above. The Suns for this lab are Ultra10s. Ed: Is looking into loadleveling/queue management software. Load balancing is for interactive logins/users. Started on it, click here for Loadleveler alternatives. Item 2: ADSM ------------ Dawn has put Compustat data out into ADSM. Hamp has contacted and copied one or two users into ADSM. Need to reconvene group to start documenting, announcing, helping users, etc. Tim S.: Suggests that ADSM needs to become a Cross-Divisional Project (CDP). To Do: (Tim) Make request to Directors for CDP for ADSM. Item 6: SP upgrade ------------------ Discussed that the Engineering School does have priority on the parallel nodes. Reviewed Tim S.'s e-mail about E-school. They get priority because they paid bulk of cost of those nodes. Tim S. hopes they "naturally" get priority because they will be biggest users. Hamp: Concerned that if they don't it'll be difficult to setup and manage user classes to give them priority. Let's hope we don't have to! He also hopes that as we shift select users from serial to parallel, we'll see a noticeable workload shift off the serial nodes. Tim S.: Noted that all the nodes (parallel & serial) are connected to switch, so we'll be able to permit us to balance workload easily. Discussion about access to the switch. Hamp: Right now, queue managing software only dispatches one job/node now, may need a flag, if load gets low, that would dispatch a second, likely short, serial job. Tim S.: Asked that we explicitly contact Mitch Rosen and ask who should be using the parallel nodes and see if we can get them going. To Do: (Ed)Volunteered to do contact Mitch. He'll also contact Steve Regean about LS-DYNA use on parallel nodes. Need to make sure it's available to SP. Tim S.: The recent SP upgrade tripled the computign power on the 16 serial nodes, so fact that 8 of 24 went to dedicated parallel nodes did not decrease available compute power, was a net increase. The eight SP parallel nodes are all the same models, CPUs, and configuration. Research Computing Support Center --------------------------------- Discussed pros/cons of using Small Hall Unix Lab as location for Research Computing Support Center and how room configuration might change. Discussed alternatives such as Wilson Hall, Clarke Hall when it expands, or Kerchof Hall. Item #4: Linux Support -------- Tim S.: Unix Systems NOT ready to offer Linux Unix Administration suppport. Hamp: Variety of issues: number of platforms, which flavors and versions of Linux? serious flaws in it make general support difficult. Only supports 16 bit UIDs. Not clear how it will resolve this, not even Red Hat. Will be an issue for new incoming students and faculty. Hamp: has reserved over 1,000 UIDs that he could re-map to other users (e.g, Linux, Next, etc.). He's used about 100 already during account consolations to the cluster. Tim S.: Researchers just want standard Unix administration suppport. Hamp: We got money for an installation server. Need to find out which Linux version the selected DCI vendor will use and hopefully use the same one. Realistically, it's highly likely to be RedHat. Tim S.: Mendel's vmware runs on RedHat (see: vmware.com) -------- Reviewed the "To Do" and "Done" table. - - - - - Meeting adjourned at 11:25, next meeting decided: May 24th 10:30 AM
Go to: ITC Research Computing Committee Home Page
by Tim F.J. Tolson