September 14, 2007

During a meeting of the computational science advisory committee, the  
subject of queuing on the clusters came up again.  There were several  
items discussed (a meeting with users, etc.); I was charged with  
bringing back some information on the existing load and users.  I'm  
sure much of this is buried in the PBS logs, but before I dig in, I  
thought I'd see what information is already available, and/or what  
kinds of tools for analyzing the logs already exist.

Questions include:

1) What are typical and extreme wait times for jobs to begin execution?

2) What are typical actual (as opposed to requested) execution  
times?  (Is there truly a continuum, or is is bi- or tri-modal?)

2) Do we know whether the HPC workshop generated any new users  
(beyond the duration of the workshop)?

3) Do we know what *kinds* of parallel jobs are being run, e.g.,  
embarrassingly parallel, home-grown or commercial code, likely to see  
great benefit from something we don't provide (LARGE memory, low- 
latency and/or high-speed interconnects)?

4) Are there any obvious candidates for the TeraGrid, whom we could  
support by helping move to this environment?

I'd really appreciate any guidance or existing data you can give me.   
Thanks.

Tom Spraggins