Charge for CAF Review, Fall 2001

Short Term Charge (September 15th):

CDF has purchased a Sun Fire 6800 computing system with 24 750 MHz SMP nodes capable of being seen as a single system image. The nodes will be upgraded to 900 MHz in September. CDF's plan is to purchase a second identical Sun Fire 6800 system in FY01 or FY02, if the performance of the first system with the CDF software is adequate.

The most immediate charge to this committee is to review the performance of this Sun system, fcdfsun2, and recommend to the offline whether or not they should proceed with the purchase of a second sun system. This should be done in the context of specific performance requirements driven by expected Run II datasets that will be available (a) for summer 2002 analysis (300pb-1) and (b) for completed Run IIa analyses (2fb-1).

fcdfsun2 should be available to users near the end of August. The committee should begin by evaluating the performance of running the offline software on fcdfsun1, a four processor 400 MHZ machine, and extrapolate to the performance of fcdfsun2. Before the completion of the report, this extrapolation should be explicitly verified on fcdfsun2.

If the committee decides to recommend against the purchase of a second Sun machine, the committee should suggest alternative uses for the funds that were earmarked for the second Sun, approximately $400K, within the central systems.

Longer Term Charge (November 15th):

The committee will review the CDF central analysis systems and recommend directions for FY02 and FY03. The committee is invited to review both the current technical choices within the CDF centralised computing model, and the model itself, and recommend directions either inside our outside the current model. The committee may consider some or all of the following:

  1. The existing and planned central analysis systems:
    fcdfsgi2, fcdfsun1, fcdfsun2, fcdflnx1.
  2. The storage and CPU resources available to these and future systems.
  3. Recommended UNIX flavors in the central systems in the future, restricted to the three possibilities IRIX, Linux and Solaris.
  4. Possible future farm-like systems for batch driven central analysis.
  5. Possible future distributed computing environments.

The report on this "long term" charge will provide important input for planning for RunIIb, which will commence late this fall.

Modified: Thu Aug 16 10:59:42 CDT 2001 Frank Würthwein