VO usage of OSG facilities:

Here are various notes:

HEP VOs:

• CMS - peak at 916k hours / week. Averaged over 800k hours / week during November 2007 (CSA07). Averaged ~700k hours/week during summer production run (July - September).
• CDF - peak at 550k hours / week. Saw about a 5x increase in CPU hours usage from Jan 2007 to Jan 2008. Still heavily FNAL-based.
• ATLAS - peak at 450k hours / week, less than CDF's 550k peak. Averaged around 250k hours / week.
• DZero shows a wide diversity of sites - no individual site has a lion's share of the work.

Question: Why does ATLAS see only about half the amount of CPU hours as CMS?

Non-HEP VOs:

• NYSgrid is a new effort which looks worth mentioning (not much usage over entire year, but ramped up from nothing to 12k hours / week in the last month)
• Engage has significant usage also (average 8k, peak 45k)
• VOs which have non-trivial amounts of usage (but perhaps not enough to highlight individually): sdss, gpn, nanohub, mipp, hypercp.

We need another VO overview page which shows what percentage of each LHC VO is represented by the T1 site.

TODO/progress:

1. Opportunistic usage plots - Initial attempts online. Need to refine site ownership definitions.
2. GridScan results graph - Online.
3. WLCG availability for CMS and ATLAS OSG sites - SAM DB?
4. Number of slots available in GIP - Can be live data.
5. Hits on the website / twiki - Contacted Anne H / Rob Q. Not available for whole year; investigating. OSG website has webalizer graphs.
6. VO usage of the OSG group by job exit code - Chris Green has built an one-off summary table
7. Downloads of OSG and VDT cache - Alain provided Excel spreadsheet.

-- BrianBockelman - 17 Jan 2008