CHEP 2000 abstract 
 

PHENIX Computing Center in Japan (CC-J) 

	Takashi Ichihara  (RIKEN, RBRC)
	Yasushi Watanabe  (RIKEN, RBRC)
	Naoki Hayashi     (RIKEN)
	Shinya Sawada     (KEK)
	Satoshi Yokkaichi (Kyoto Univ.)
        Atsushi Taketani  (RIKEN)
	Yuji Goto         (RIKEN)
	Hideto En'yo      (Kyoto Univ., RBRC )
	Hideki Hamagaki   (CNS, Univ. of Tokyo)
                         
Abstract

  PHENIX Computing Center in Japan (PHENIX CC-J) is now under construction
at the RIKEN Wako campus over a three year period, beginning in April 1999. 
The CC-J is intended as the principle site of computing for PHENIX simulation,
a regional PHENIX Asian computing center, and as a center for the analysis 
of RHIC spin physics. The planned computing capacity will be sufficient to 
meet the bulk of the simulation needs of PHENIX. The planned robotic storage
capability will permit efficient micro-DST production as a regional
computing center.

  The current and planned size of the CC-J system is as follows.

		      Oct.1999   Dec.1999  Dec.2000  Dec.2001
 CPU farms (SPECint95)  1360       2400      5900      10700
 Tape Storage (TB)       100        100       100        100
 Disk Storage (TB)       2.6          5        10         15
 Tape I/O (MB/s)          45         68        90        112

   These planned capabilities of the CC-J are in concert with the research 
plans of RIKEN and RIKEN BNL Research Center, which initiated the RHIC 
spin program and which plan to expand their involvement in the entire
PHENIX program.

   High Performance Storage System (HPSS) was installed in April 1999 
and more than 50 MB/s sustained transfer rate is obtained by pftp
between HPSS clients and HPSS servers over the Gigabit Ethernet. The 
utilization of the Jumbo-frame (9 KB MTU) Gigabit Ethernet reduces 
the CPU overhead for the frame analyses thus the high performance of 
the data transfer has been achieved.

   Linux pc's with Intel Pentium II/III are used for CPU farms.  The 
NFSv3 patch for the Linux latest kernel significantly improved the NFS 
performance, especially for the NFS write operation.  Portable Batch
System (PBS) are used for Linux pc farms. To access AFS cell stably,
a local mirroring of the AFS contents is carried out daily.
 
   Further information of the CC-J can be found at the URL of
http://ccjsun.riken.go.jp/ccj/ .

-------------------------------------------------------------------------
Key Topics
  Commodity Hardware and Software and Integration in Farm and Large Systems

Kew words
  HPSS, Gigabit Ethernet, Jumbo frame, Linux CPU farms, NFSv3 kernel patch, 
  PBS, AFS, Objectivity/DB, RHIC, PHENIX