The BELLE Computing Facility
Paper: 217
Session: F (talk)
Speaker: Itoh, Ryosuke, KEK, Tsukuba
Keywords: hierarchical storage management, large systems, massive parallel systems
The BELLE Computing Facility
H.Fujii, R.Itoh, N.Katayama, A.Miyamoto, T.Nozaki, Y.Takaiwa,
T.Tsuboyama, M.Yamauchi, Y.Yasu
Physics Division, National Laboratory for High Energy Physics (KEK), 1-1 Oho, Tsukuba, Ibaraki 305, Japan
S.Karita, S.Kawabata, A.Manabe, S.Sasaki, Y.Watase
Computer Center, National Laboratory for High Energy Physics (KEK), 1-1 Oho, Tsukuba, Ibaraki 305, Japan
Y.Watanabe
Department of Physics, Tokyo Institute of Technology, 2-12-1
Ohokayama, Meguro, Tokyo 152, Japan
The computing environment for the data analysis at BELLE experiment
(the B-factory experiment at KEK) is described.
The system was designed to have the processing power of
50000 SPECint92 and the storage capability of 30TB/year to process
the data from the data acquisition system fed at 15MB/sec.
The main computing power is provided by 7 units
of the SUN UltraEnterprise server with 28 CPU's. These servers are
connected together by the Fujitsu's high speed network called AP-net
with MSI GEN-5 RAID disks of 3.7TB. Some part of the disks are migrated
with the SONY DTF-based tape library system using the OSM.
For the raw data storage, the SONY DIR-1000 drives are used with
the library system.
The program development for the servers are done on many desktop
SUN workstations and work-group servers scattered in KEK.
These systems are connected to the central facility
through site-wide ATM network. The network is also connected to
the Japan-wide ATM network cluster (SINET-ATM) as well as to the gateways to
foreign countries for outside users.
The entire system is operated under the DCE environment and the DFS is
used to manage file systems. This enables to have the unified directory
structure from all the workstations/severs and to place the
replica of a data file on local machines without taking special care.
The resource management of the system is done using the LSF so that the
CPU power on scattered workstations and servers can be efficiently utilized.
The system is now being installed at KEK and the preparation for the
operation is in progress. The operation of the system will start
in January 1997.
The detail of the system will be covered at the conference
with the report of the initial performance.