GDB 11th March 2009

Notes By Pete Gronbech

Dave Kelsey Security Docs.

Responsibility on the VO manager In order to comply with the new rules VO managers will have to keep extra details such as dates when users registered and signed (and which version of the AUP was in force at the time), along with expiry dates. VOMSRS apparently has the ability to record this sort of thing but VOMS (such as run at Manchester) does not. It may be necessary for the VO manager to setup their own records!

Accounting Privacy issues Should be ok if we keep data within EU as all countries should have the same privacy laws? If you wanted to see what ‘our’ users were doing how can you do it? Often UK people have cern certificates and vice versa. The nationality, institute association are not stored.

Steve Traylen talk very interesting. Reporting on installed capacity, new glue schema and yaim to cope. New HEPSPEC2006 values can be added now and old spec int left until apel changed to accept new hepspec values. At that time the specint values should be calculated from the hepspec2006 values rather than old estimates.

Middleware news

Lots of talk about moving to SL5. The WN is or will be available real soon now. Some sites say they will only move once an SL5 UI is also available.

Disaster planning; see analysis of the recent fire in ASGC Their UPS batteries caught fire, but the mess in the whole room was awful. They very systematically cleaned up and got back on line pretty quickly. We should have a disaster recover plan!

Lots of talks about the CREAM ce. Once all the criteria

https://twiki.cern.ch/twiki/bin/view/LCG/LCGCEtoCREAMCETransition 

are met sites will be asked to transition from lcg-ce to cream, therefore sites are asked to install cream ce’s now in order to get practice and familiarity with it. Can either do it now or wait till the end of this month for the next version.

All expts now keen to go to SL5 /64bit with 32bit compat libs. Questions about increased memory requirements caused by moving to 64bit. Question of jobs getting killed when it goes over vmem limit or physical mem limit.

Want sites to provide both in parallel rather than big switch?

SCAS Memory leaks. This is because there is one process rather than one per process. Some fixed. Some from globus, being chased. Restart trick is borrowed from apache. Error rate should be 0.02%, measured was 0.2% ML should have realised this was something else. IO lib was released in wrong order. New testing have 1 error in/4 million. Have only tested with 10 wn’s need larger site testing.a