1. 1. July 2005
   2. :00 - 17:30

Attendees ---------------------------------

ASCC

Di Qing

BNL

FNAL

Jon

FZK

Jens Rehn

IN2P3

INFN

Luca, Michele Michelotto, Daniele Bonacorsi

SARA

Ron

NGDF

Lars

TRIUMF

Reda

PIC

Gonzalo, Jose Hernandez

RAL

Andrew Sansum, Martin, Derek

CERN

Jamie Shiers, James Casey, Gavin Mc Cance, Patrica Mendez Lorenzo, Sophie Lemaitre, Vlado Bahyl, Roberto

DESY

Michael Ernst, Patrick Fuhrmann

Expts

Lassi Tuura, Nick Brook, tim Barrass

Subject -------------------------------------

CERN Overview Not as far as we want, but now we've got the castor2 major problems solved - rate is back up .

Need to work on ASCC

RAL

Problem with gridftp servers hanging from time to time.

TRIUMF

Pool node hangs at 99% CPU on one transfer and accepts no more connections.. Are other people seeing that ?

Michael: can you send us the logs for that node.

SARA

Ron : We saw a lot of put entries - restarting didn't really help - we had to clean up the postgres database. Also we've had a pool fill up - seems we saw some other pool nodes freeze - didn't get enough info from the log files, but the debug limit is up again.

This morning we had a pool node that crashed twice since it ran out of memory when we tried to get up to 150MB/s. Now throttled back down.

Michael

What is running out of memory - the JVM?

Ron

Yes.

Lassi : See some problems with timeouts at RAL and CNAF

Derek

We allocated 12 TB, 10TB was already allocated.

Michael

10 streams per transfer - on average 8 transfers per pool nodes.

Jon

At FNAL we see 20 streams per transfer - 5 per pool with 2 pool partitions per system.

IN2P3

We had problems for the last two days - srmcp worked fine, but glite-url-copy works.

FZK

Jos: Now have transfers - 3 pool nodes only connected due to network problems at the FZK side.

NDGF

PIC

gonzalo: saw many problems yesterday - this morning started again - running smoothly.

James

Tuning might be different.

ASCC

Di: Network bandwidth very low. - only one machine being used - gridftp. Need to get SRM working.

Michael

We've seen some problems with individual nodes at the castorgridsc cluster - are these resolved.

James

Can we get a path at DESY we can write to - faster to fix it ourselves rather than round-tripping.

Jamie

won't talk about service phase now - we are focussing on throughput phase. But need to prepare for GDB - sites can co-ordinate with Jeremy/Jamie/Kors. One issue is that the sites will come back with their sample jobs for validation.

James

Load generator can be run from either end. If you want to run your own, feel free - fill in the table in SCThreeThroughputHowto to say if you want to run it yourself.

Jamie

No meeting next week due to the GDB - people can dial into that.