- Working group meetings notes:
- Riccardo Di Maria will take care of it starting from today.
- Some possible next topics (not in order) and suggestions welcome:
- DOMA in IRISHEP and Analysis facility, serviceX etc. etc.
- Dedicated session dedicated to network scheduling, and how this fits into Rucio & FTS: SENSE, AutoGole, Rucio, FTS, Escape, others?
- Network vs caching cost
- Tape recall modeling study in CMS
- Caching study of the SoCal cache using MONIT in formation from Xroot servers and HTCondor classAds
- Data integrity in XCache
Minutes
- Andrey Zarochentsev presented XCache on RU-DataLake deployment and testing with ATLAS tools
- JINR SE dCache: reading through XCache & direct access
- goal was to check:
- potential gain for 'small' diskless sites which are 5-10 ms away from SE
- HC instrumentation
- monitoring systems
- summary:
- XCache is configured and works well
- automatic configuration in AGIS with XCache is checked
- synthetic tests show 95% gain in time for 100 repetitions of a file copy
- HC tests show 30% gain in time for “copy2scratch” and 12% gain in time for “Direct access”
- details shown on XCache monitoring and on PerfSonar
- Xavi commented that this work can be exported to other regions
- Riccardo Di Maria presented status of the ESCAPE CERN XCache
- simple CERN VM-based setup
- CMS-INFN XCache experience reproduced
- CERN vanilla XCache deployed
- Disk Caching Proxy (DCP) cluster deployed
- open discussion/questions wrt possible issues faced within DCP implementation)
- Next Steps - To Do List:
- implement monitoring
- integration with other storages besides eulake
- investigate other protocols such as http
- XCache stress test using HammerCloud mainly to investigate data corruption
There are minutes attached to this event.
Show them.
