wiki:developers-meeting-20160209
Last modified 3 years ago Last modified on 02/09/16 15:05:10

dCache Tier I meeting MONTH DATE, 2013

[part of a series of meetings]

Present

dCache.org(Paul), IN2P3(), Sara(), Triumf(), BNL(), NDGF(), PIC(Marc), KIT(Samuel), Fermi(), CERN()

Agenda

(see box on the other side)

Site reports

PIC

Marc reported that everything is more or less OK.

ATLAS staging requests

They are currently suffering from a problem with their tape system. ATLAS are currently reprocessing all data over a three year period, so triggering massive restaging activity: PIC currently has a queue of some 40,000 movers. Because PIC has only 8 drives, shared between ATLAS and CMS, staging these files is taking longer than the timeout period, so dCache is killing the stage requests. Unfortunately the enstore script has a bug where cancelling a stage request does not work, resulting in "background" activity beyond what dCache is controlling.

The Fermi guys said they have never seen this; however, since Fermilab have a 7 day timeout on staging requests, they never see this problem.

PIC has developed a patch to fix this problem. It isn't deployed yet.

The rest is OK.

Splitting SRM

Marc is preparing for a downtime towards the end of March. He is intending to split the SRM into separate endpoints: splitting the databases.

PosgreSQL v9.5

Marc also asked if there is any known problem with running PostgreSQL v9.5. Paul said no: there's even some optimisation available with 2.14, but unfortunately not 2.13.

KIT

Samuel reported that everything is running fine at KIT.

PowerPC

Samuel asked if there was any experience with running dCache on a PowerPC little endian machine. Paul replied that some time ago (at least 5 years) there was an unsuccessful attempt to deploy dCache on a PowerPC. This was unsuccessful because, at that time, dCache used JGlobus 1.8, which depended on an old TLS library that didn't work on IBM's JVM for PowerPC. This should not be a problem as 2.10 to 2.13 uses jGlobus 2.0 and 2.14 uses CaNL. Both these libraries use the JVM-supplied TLS implementation, which should be compatible with the JVM-supplied implementation.

NFS deletion logging

Samuel also asked about the current status of supporting NFS deletion notification in billing. Paul said this is in-progress. He couldn't give a firm estimate of when this would be available, but said he would bring it up during the developer's meeting.

Support tickets for discussion

[Items are added here automagically]

DTNM

Same time, next week.