wiki:developers-meeting-20160112
Last modified 3 years ago Last modified on 01/12/16 14:51:16

dCache Tier I meeting January 12, 2015

[part of a series of meetings]

Present

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

Agenda

(see box on the other side)

Site reports

PIC

Things are OK now

There are two disks that have failed --> IOException. Small pool for Tier-2

If you mark file as broken and a pool is restarted then the file will be OK.

From admin interface you can remove the "broken" bit.

New heardware for pool; deploying with CentOS 7.

NDGF

Had an "Interesting" Christmas.

Alice pools dying under heavy load. These 80 TB ZFS; this is very slow for dCache. Alice fills these disk with small files. Have got the pool start-up from 6--8 hours down to 1 hour.

A lot of different things. Ulf guesses the average file size 50--100 MiB.

ZFS IOPS depends on the slowest disk in the system. Alice does a lot of xrootd vector reads; most less than 1 kiB in size. The block size is 256 kiB. Checksumming means this is very bad performance.

At DESY we tried Linux ZFS on Linux the performance was unpredictable and after a week

Planning to switch from ZFS to XFS during down.

News section of webpage needs updating.

Book: 2.13 2.14 links are broken.

KIT

Everything is running just fine.

Support tickets for discussion

[Items are added here automagically]

DTNM

Same time, next week.