Home » Alice » Alice Computing » Computing Meeting, Wednesday, November 21
|Minutes of the ALICE Computing Meeting, Wednesday, November 21 [message #5447 is a reply to message #5442]
||Wed, 21 November 2007 14:55
Registered: March 2007
1) PROOF cluster|
PROOF cluster is up and working. Login problem (machine respond delay) doesn't exist anymore probably due to not running NIS cache. Marian was satisfied with a proposition to let 2 CPUs for PROOF and 6 for other jobs on batch machines. Victor suggested that machines have to be restarted then.
2) PROOF tutorial
It was agreed that PROOF tutorial should take place on Friday 23rd of November at 10.00. The names of participants should be collected beforehand. Discussion arouse about who should be on the list. Anton and Marian proposed that full "analysis list" of local people must be added. Everyone on the list has to be in administrators' reach if any problems appear. The participants will get a login access as PROOF users. The idea to take Alice pool accounts and use them as test users accounts was rejected. Peter Malzacher offered a convincing argument that administrators need an opportunity to find out who did what within a reasonable time scale. No anonymous accounts should be allowed. Thus, if e.g. Marian needs a couple of accounts, he must apply for more accounts which should look like marian01, marian02, etc.
3) Silvia's grid jobs
29 of 30 grid jobs finished successfully. The one that failed had to copy something from CERN, so maybe GSI is not to blame. Timings improved and start to make sense. Silvia asked, how much jobs from now on she has to submit and how often. Kilian proposed automated mechanism of doing it.
4) Ana's grid jobs
Ana's grid jobs tend to be unlucky.
5) Misha reported on the progress of speed tests. Results show that processing of 1 million events in 1,50,250, and 1000 files goes with almost the same speed of 15 MB/s. With 10000 files it's 11 MB/s.
The preliminary results of testing of reading files with pythia events give a hint that possibly the internal structure of the trees affect reading speed. As for files with pythia events reading speed is higher.
The test on data stored at Lustre cluster is not consistent because Lustre cluster was used by others at that time. But the results of the test show that while running parallel jobs, the events/s parameter scales with a number of jobs.
6) Full-dress rehearsal
Weekly meetings are organized, but there is an attendance problem. The idea behind full-dress rehearsal is discussed: whether it means quality assurance, or rehearsal of full chain of processes.
7) List of files transferred to GSI from CERN
Kilian assured that a website together with a WIKI page where the lists will be stored are ready. XML files with filenames are available.
xrootd POSIX interface
Kilian reported that compilation of the POSIX interface for xrootd developed in Dresden has failed on sarge (release of Debian). So he will try to compile on etch release.
[Updated on: Wed, 21 November 2007 15:23]
Report message to a moderator
Current Time: Wed Dec 08 07:35:10 CET 2021
Total time taken to generate the page: 0.02232 seconds