Minutes SLAC-ALTAS TDAQ Meeting 14-Dec-2006 ------------------------------------------- Present: Matthias, Rainer, Sarah, Su Dong, Amedeo, Steffen Sarah: LST finished Monday morning. Three tests related to dbproxy: a) dbstresser 600 processes directly to the database vs through proxy. MySql direct access didn't complete until after 172 s, 250 s for the last. First proxy at 15 s with dbproxy, last one in 490 s. (?) This is not what you would expect. 200 MB of data. Not great control over other network access. Supposedly exclusive access to CPUs. Amedeo suggest to make a histogram and see the distribution of the nodes. b) Test of L2 with Haimo Configured up to ~240 nodes. Looked at a couple different configurations. First case, 12 proxies, no top level; 2nd direct MySql; 3rd SQLite files local; 4th connect to Oracle. Only 15 s out of 70 s could be attributed to the proxy. Observed 2 s connect per proxy. c) Test EF with Serge Same 240 number of nodes. 1PT, 2PT or 4PT per nodes. Also used 12 proxies. Will circulate a draft of the results. Preloaded the proxies. Comparing to not doing that, the tails were chopped off. Not sure how much data he transferred. His configuration took 6 minutes. He never saw worse performance with the proxies, often no change in performance. Pre-cached proxies seemed better. Over 1000 connections made to the Oracle server. Had to be hacked to allow that. It was noted that without proxy it would have been 100k connections. People were pleased that the tails went away with preloaded proxies. A little surprise that we didn't see more improvement with the proxies, though. Consensus that we need to test in a more constrained environment. Open meeting with discussion of LST tests. Request late January. Amedeo asking about CORAL atomicity. For unknown reasons MySQL and SQLite versions of the geometry are incorrect, while Oracle is correct. Amedeo: Talked to Andy Hanushevsky. Some discussion on Andy's generic advice, but will have to follow up with more concrete issues.