CCRC08
2. 6. - 5. 6. 2008 Analysis Latency Tests
Exercises consist of subscribing to a data set and running a sample analysis job with CRAB over all its files.
Monday and Tuesday CSCS had a number of problems due to a middleware upgrade of the CE and nodes which despite expectations caused a lot of jobs to abort. But I was able to download a dataset from PIC at a speed of ~90MB/s.
Tuesday evening and night was ok, but allowing only one cycle for the following data set:
Dataset name:
/Njet_5j_400_5600-alpgen/CMSSW_1_6_7-CSA07-1206675842/RECO from T1_PIC
Time |
Time delta |
Comment |
15:00 |
0 |
Dataset subscription approved |
15:25 |
0:25 |
download beginning |
20:45 |
5:45 |
dataset completely on site (281 Files from T1_ES_PIC, 1.4TB, 0 Errors, avg. 69.1 MB/s) |
21:40 |
6:40 |
DBS still only shows 41 files at CSCS |
21:43 |
6:43 |
DBS shows whole dataset to be on site |
21:45 |
6:45 |
crab -submit |
21:49 |
6:49 |
all 29 Jobs running at T2_CH_CSCS |
04:34 |
13:34 |
last job finished (no Errors) |
Wednesday was again bad luck, when FZK's network had severe problems, causing all transfers for
/WW_incl/CMSSW_1_6_7-CSA07-1196178448/RECO
to fail.
FZK recovered sometime in the afternoon, but the CSCS PhEDEx download agent got stuck because of blocking of a
glite-transfer-query
and a
glite-transfer-submit
request, which I only noticed on Thursday morning.
PhEDEx transfers from FZK began to trickle in by 13:30h, but the cluster is now completely filled with user jobs, which will add considerable latency to the eventual CRAB run.
--
DerekFeichtinger - 04 Jun 2008