Speaker
James Botts
(L)
Description
PDSF (Parallel Distributed Systems Facility) has been in continuous operation at NERSC since 1996 on dedicated and ever changing hardware, supporting a broad user base in the high energy physics community. We will describe recent and ongoing changes in the underlying architecture of PDSF. We are moving to a model where the PDSF cluster will consist of dedicated front end servers and a set of compute nodes drawn from a backend cluster, unified by a batch system. User storage will be migrated from a dedicated GPFS cluster to a site wide NERSC Global File system, facilitating the use of other NERSC clusters which have a faster and lower latency interconnect for parallel jobs. These changes should help minimize downtime when the cluster is moved to a new building at Lawrence Berkeley Laboratory in 2014/15.
Author
James Botts
(L)
Co-authors
Iwona Sakrejda
Larry Pezzaglia
(LBNL)
