[Japanese]
Research Center for Nuclear Physics (RCNP), Osaka University, is managing a vector-parallel supercomputer NEC SX-5.
This system is being managed in cooperation with Cybermedia Center (CMC), Institute of Laser Engineering (ILE), and RCNP. Present machine NEC SX-5 was placed on 26 December 2000, and started official run from 5th January 2001.
NEC SX-5 is a vector-parallel supercomputer, and successor of NEC SX-4, which had also been managed by the three centers from January 1997 until December 2000.
The quolification for RCNP supercomputer system is the same as for the generic unix system of RCNP. They can be used by researchers in theoretical or experimental nuclear physics including interdisciplinary areas. For details of the quolitification, please access here .
One must get an account of generic computer system before
applying to supercomputer system.
For applications to these systems, please follow the guide in
the page of Computer and Network Group.
The structure of the system is described in Users' guide
Among the whole system (16 processors x 8 nodes),
RCNP user can use the following parts:
the node managed by RCNP (RCNP node, 1 node),
and about 1/6 resources of the nodes managed by CMC
(CMC nodes, 6 nodes).
RCNP node is for single processor computations,
and CMC nodes are for parallel computations.
The above User's guide mainly describes how to use the RCNP node.
If you use CMC nodes, please follow the guide in
the page of CMC .
Users' guide (3): parallel
computaitons for RCNP users is now under construction.
Please tune your program towards high performance computaiton suited to vector-(parallel) machine. If your program has low vectorization rate, there is no merit to use the supercomputer compared to workstations.
The following pages may give hints for the tuning for vector, and parallel computations (now under construction):
For the algorithms not suited for vector processors, you are recommended to use a computation server. The computation sever can be used from generic unix system (senri) by submitting batch jobs via NQS (network queue system). Details are explained in: