Pre-GDB - Grid Storage Services

Tuesday 11 Nov 2008, 10:00 → 18:20 Europe/Zurich

IT Auditorium (CERN)

Flavia Donno (CERN)

In depth discussions with the developers, who will give configuration and management hints to site admins and site admins will have an opportunity to list their problems and express their requests. EVO Conference: http://evo.caltech.edu/evoGate/koala.jnlp?meeting=e8eteDv8vuava9IaasI8 EVO Phone Bridge: 599586

Minutes pre-GDB_Minutes.txt

10:00 → 12:30 Tier-2s

This session will be dedicated to Tier-2s running DPM and StoRM.

10:00 Introduction and experiment requirements

Speaker: Dr Flavia Donno (CERN)

Slides pre-GDB_Storage.pdf pre-GDB_Storage.ppt

10:15 FTS: an update

Speaker: Akos Frohner (CERN)

Slides PreGDB-IT-DM-FTS-2008-11-11.pdf PreGDB-IT-DM-FTS-2008-11-11.ppt

Items arising from the discussion after the presentation: Full listing of channel parameters: https://savannah.cern.ch/bugs/?43819 Checksums support: https://savannah.cern.ch/bugs/?43825 [FTM] cleanup and publishing of transfer logs: https://savannah.cern.ch/bugs/?43837

10:30 GFAL and lcg_util

Speaker: Remi Mollon (CERN)

Slides PreGDB-IT-DM-clients-2008-11-11.pdf PreGDB-IT-DM-clients-2008-11-11.ppt

10:45 DPM status and plans

An update on DPM status, plans and the currently recommended release.

Speaker: David Smith (CERN)

Slides PreGDB-2008-11-11_DPM-SE.pdf PreGDB-2008-11-11_DPM-SE.ppt

11:05 DPM configuration and discussions - question from the sites

* user quotas * tools for performing common administration tasks * access control for spaces * more advanced pool selection mechanism for DPM * improved logging (centralised) * tools for checking SE-LFC synchronisation * nagios style alarms * Are all of the current SEs set up properly in order to optimally deal with local user analysis?

Speaker: All

Items arising from the discussion: Addition of more filesystem selection methods https://savannah.cern.ch/bugs/?43932 Tool to balance filesystems https://savannah.cern.ch/bugs/?43931

11:25 StoRM status and new release

Speaker: Luca Magnoni (CNAF)

Slides StoRM_status.pdf

11:55 StoRM configuration and discussions - question from the sites

* Some sites are testing StoRM now. Some sites are using GPFS and others Lustre. Can we expect exactly the same functionality from both types of SEs? * access control for storage area * gridFTP load balancer

Speaker: All

Slides

• questions-GDB-11-11-storage.pdf
• questions-GDB-LIP
• StoRM_configuration.pdf

12:30 → 14:00 Lunch break

14:00 → 17:00 dCache for Tier-1s and Tier2s

This session will cover dCache specific issues and discussions

14:00 dCache status and plans

An update on dCache releases recommended for Tier2-s

Speaker: Dr Patrick Fuhrmann (DESY)

14:20 Installation and configuration hints for Tier-2s running dCache

Advise on how to install, configure and run a dCache installation at a Tier-2

Speaker: Dr Patrick Fuhrmann (DESY)

Slides gerd-pre-gdb-081111.pdf.pdf pregdb-nov-dcache-11112008.pdf

14:50 Coffee break

15:05 dCache configuration for Tier-1s

Speaker: Dr Patrick Fuhrmann (DESY)

15:45 dCache - questions from the sites

1. Splitting SRM instances. Is it feaseable ? 2. Configuration of gsiftp doors 3. Hardware setup at sites (32 or 64bits? How much memory and where? ...) and recommended software packages to use (Java version, DB, etc.) 4. What versions on the head nodes and what versions on the pool. 5. Avoiding high load on the PNFS node(busy threads in PnfsManager - long queue) 6. Limit the number of requests of a certain type (put request) globally or per user. 7. "We currently experience a long-standing problem of storage availability at CERN that I think is worth discussing (again) at the preGDB. One disk server of T0D1 service class has now been down for 2 weeks. We have a few thousand files on there that are inaccessible :((( It is a real burden to figure out, even if we could get the list (which we don;t have). How do sites envisage to face such problems?" - Philippe Charpentier 8. Managing databases: Putrequest and Putrequesthistory tables becoming big.Should "[Vacuum] analyze" being performed? How often ? 9. What should clients be careful using ? What are the most expensive calls ? smrls -l 10. How big should be the directories in order not to run into pnfs slowness ? 11. What is the advice for dCache sites regarding migrating to Chimera? It seems as if no one has moved yet. 12. Are all of the current SEs set up properly in order to optimally deal with local user analysis? 13. Implementations for 64 bits architetures 14. Is there a plan for a user-friendly tool for dCache installation 15. What is the status of the dCache dynamic information providers ? 16. The "fast" PNFS implementation 17. Poor ability of some dCache componentes to scale 18. No automation for disaster recovery. Many disaster recovery-related actions have to be done by hand, which is not our target. dCache somewhat hinders the possibility of deploying high-availability solutions - though allows easy rapid-recovery solutions. 19. PIC: Monitoring of our system as a whole is not at the level we would like it to be. Performance bottlenecks are difficult to find. We are working to improve this. 20. IN2P3: Massive prestaging from tape system to dCache are not efficient enough: when pretaging through SRM, dCache sends the prestaging requests in little packets (say 10) simultaneously to HPSS. It causes a lot of inefficiency in the tape mounting and unmounting. This is the big bottleneck we experience in the prestaging exercises.