Tech Meeting Minutes 20110303
Present: Graeme, Wahid, Andy, Sam, Stuart, Mark
Simpsons Trivia: What almost caused the nuclear meltdown at the Springfield Nuclear Power Station?
|Table of contents|
- No one present. Noted worrying build-up in number of open tickets against the site. Mark will chase this up.
- Two tickets related to information publishing (also Glasgow and Durham have been ticketed). Tickets are very non-specific and so have been marked as 'waiting for reply' awaiting exact details of the problem.
- Two related incidents last week:
- Unscheduled power loss on m/w node rack. Supply tripped and there is no backup. Powered up ok and back within a few hours.
- CREAM CE's odd behaviour was tracked down to MySQL DB corruption after the power loss - this caused high load from MySQL trying to query broken tables.
- Will drop and recreate DB. ATLAS are ok with this as pilots run independently, but Graeme will check factory after Andy does this.
- Wahid reported good results from 2 HC tests. These are D3PD maker tests, ran at good CPU efficiency.
- Agreed that testing should continue, but Graeme thinks this is good enough to enable PD2P at ECDF. ACTION.
- Ticket 66929 should be closed (actually relates to a failing CE at IC).
- Mark reported network tests went very well on Tuesday. Bypass of security module resulted in a smooth 2.4Gb/s in and outbound, which is 1/3 of the Clydenet bandwidth. See http://scotgrid.blogspot.com/2011/03/wide-area-wonder.html.
- We now have no security enforcement from the campus routers - need to add to our own security to compensate.
- Very poor network connectivity to Edinburgh has been noticed for some months (very high RTT) after a carrier upgrade.
- No resiliance on our link to clydenet at the moment.
- Want to re-run tests during a peak load time to see the effects.
- Stuart has been trying to pull data from LB servers to monitor pheno performance.
- Problem with bupdater on one CREAM CE, but not others. CREAM developers are trying to improve blah, because it sucks.
- Sam has been developing an automated test system for 2GB files moving between GLA, MANC and BNL, SARA, RAL.
- BNL rates are still very poor.
- SARA rates now x8 better than sonar.
- RAL results are excellent (40MB/s on a single stream).
- Graeme: aim is to commission Glasgow as a T2D soon.
- Next meeting: Feb 24, http://indico.cern.ch/conferenceDisplay.py?confId=129019. (Graeme on holiday, Mark to chair.)