NUC 11 Dec 2013

Yves Kemp; Kai Leffhalm; David South; Hartmut Stadie; Michael Steder; Shaojun Lu; Andreas Gellrich

problems:
- 1 user requesting 15 GB ram per job; resulting in ~950 jobs on ~1600 cores; low throughput
- CVMFS: problems at CERN affecting our ATLAS and CMS clients
- CVMFS-over-NFS served old file once, fixed by remount
- only one NAF-1 login machine per experiment still running due to old hardware
- /nfs/dust mount not visible for users due to automounter
- NAF-2: ILC WGS unresponsive, needed reboot because of memory leak by one user
- NAF-2: CMS WGS unresponsive, /tmp was filled up
Monte Carlo in Munich hosted on NAF 2.0
- 90 school accounts, AFS group directory, 5 WGS for interactive login
- worked well, easy setup
- if you plan schools, contact NAF team
since 3 Dec BELLE dCache mounted on BIRD/NAF-2 on all SL6 nodes, looks promising
LHCb discontinued on NAF-1 in Zeuthen, Tier-2 remains; will move to HH for user analysis
WN batch migration; 500 cores coming weekend, 500 cores end of month

id	who	description	status/comments
1007-2	NAF	report on help desk tickets	shown
1210-2	NAF 2.0	offer better monitoring for queue usage (per user, waiting time, run time, etc)	tested Splunk, currently investigating other projects; use http://bird.desy.de/status
1210-3	NAF 2.0 & experiments	reconsider queue setup, scheduling, dynamic or configurable queue limits
1302-2	NAF admins	clearly document what scratch space means and policy for Lustre and SONAS

who

description

status/comments