Computer Clusters for CMS in Russia.
NPI MSU and ITEP Pariticipation in HLT Runs.

Few computer clusters (in NPI MSU, ITEP, JINR, IHEP) have been created. CMS software (fortran and C++) have been installed. Work on the possibility to have common file, batch and archive system began. Objectivity DB was installed and work on data base replication and mirroring will begin soon in collaboration with CERN, USA, Italy, Pakistan.

In Febuary two clusters NPI MSU and ITEP participated in the High Level Trigger run which was done for the muon group in collaboration with CERN, USA, Italy, Finland. From the NPI MSU side 6 Intel PC (3 Pentium II/350 MHz, 2 Celeron 400MHz and 2xPentium Pro 200 MHz with 128Mb RAM and 17Gb disk) were occupied with production. In ITEP cluster consists of 9 Pentium II 400MHz with 128Mb RAM and 63Gb of disk space was available for this work.

We shared with Helsinki university the generation and passing through CMSIM 50000 of pile-up events with two energetic muons. Sample of 25000 pile-up events (our part) was generated with PYTHIA and passed through CMSIM for producing HITS for all detectors. Events were generated by the pieces of the 500 events. The mean time of the 500 events job was 80000 sec including both PYTHIA and CMSIM generation. The size of output file is around 1 Gb (2Mb/event).

Obtained files were transferred via internet to CERN using network channel Moscow-DESY-CERN. The bandwidth of channel is 512Kbit/s. During one week it was completely occupied via production but since 24.02 it was limited to 380Kbit/s for our studies. The network allows to transfer around 8Gb of zebra files per day on using scp in compression mode. Few files (till 8) were transferred at one time. The stable work of channel was observed for up to 6 files transferring. The larger number of files lead to increasing the transferring time because of overloading of bandwidth. The mean transferring time to CERN of 1Gb file is 20 hours under condition of normal load of channel. The part of production statistics (day by day and integral job running and transmission) made in NPI MSU is shown on the figure. At present our production capability in CPU for this two clusters is two times higher than our capability to transfer all information to CERN via Moscow-DESY-CERN channel. Using 15 PC we can generate near 8000 events of this type per day and get 16Gb of data and only 8Gb can be transferred to CERN.

In future production we intend to increase the number of participating computer clusters using the facilities of other institutions (Dubna, Protivino, LPI etc).

The process of events generation at NPI MSU and ITEP can be illustrated by the total statistics on runs and statistics on runs in details

Submitted by O.Kodolova