Structure of CAF Review, Fall 2001

Offline management notes that it started this review to involve the CDF collaboration in the planning process for the fiscal year 2002/2003 computing budget, and to receive guidance from a larger crossection of CDF.

The committee notes that there appears to be significant discontent by the CDF collaboration with their experience as users of the central analysis facilities (CAF) at CDF.

However, at the outset of the review it is far from clear why CAF doesn't live up to user expectations. We have thus decided to divide the review into the three main themes below.

The committee notes that Data Handling aspects are explicitly not within the scope of this review. A seperate review is being organized by the spokes people of CDF that addresses data handling issues. There are presently 7TB of disk space installed in CAF, and another 30TB purchased and to be installed soon. For the purpose of this review we will assume that there will either be sufficient disk space available to have physics analysis data (PAD) on disk, or that data handling issues will be resolved such that staging is not a significant limitation in CAF performance. We are fully aware that being wrong in this assumption may invalidate any recommendations we will arrive at!!!

• Benchmarking:
  Establish a performance matrix for a set of standard CDF Run II applications on a set of standard platforms. This includes CPU as well as I/O benchmarks (Details).
• Review of the expected Usage Profile:
  The CAF was originally speced (CDF Note 4072, CDF Note 4100) in 1997 based on Run I experience (summary of 4100,4072,4718,5565). We will review those specs both to see if they have been met and to see if they are sufficient in first place given possible changes in the needs of the physics groups beyond the 1997 expectations.
• Review of the Computing Model:
  Review the CAF computing model, outline possible alternatives, explore advantages/disadvantages of different models. CAF hardware architecture is described in CDF note 4707 as well as a CHEP2000 paper.