Reprocessing H1 Data on the IN2P3 computer farm in 1997

Paper: 435
Session: F (talk)
Speaker: Dagoret-Campagne, Sylvie, LPNHE University Paris
Keywords: networking, network monitoring, network performance, parallelization

Large Reprocessing H1 Data on the IN2P3 computer farm in 1997

Sylvie Dagoret-Campagne
LPNHE University Paris VI-VII

IN2P3-H1 group and H1 Collaboration

With the increase of Luminosity and the increase of duty cycle of the HERA
accelerator in 1996, The H1 colaboration must increase its computing
power for reprocessing all its data, about 90*10^6 of events (5 TBytes)
acquired this year.
The chosen solution is to undertake the reprocessing outside Desy center,
where it was carried out up to now, apart running periods,
on a multi-processor computer (Unix like SGI Challenge).
I will present the software which has been develloped in collaboration with
the CCIN2P3 for reprocessing 30^6 events during a 3 months period on
a Unix workstation farm of the computing center of the IN2P3.
Reconstruction task are running on parallel on this distributed achitecture
consisting of about 15 HP and IBM-RISC machines under the control of a batch
system (BQS) which shares resources between the users.
The essential point of this developement rely on data servers (based on
TCP/IP protocol of Unix sockets), used for event distribution (to) and
collection (from) the workstations distributed on the network.
The procedure, the performance and the results of this reprocessing will be