Speaker
G R. Moloney
Description
We have developed and deployed a data grid for the processing of data
from the Belle experiment, and for the production of simulated Belle
data. The Belle Analysis Data Grid brings together compute and storage
resources across five separate partners in Australia, and the
Computing Research Centre at the KEK laboratory in Tsukuba, Japan.
The data processing resouces are general purpose, shared use, compute
clusters at the Universities of Melbourne and Sydney, the Australian
Partnership for Advanced Computing (APAC), the Victorian Partnership
for Advanced Computing (VPAC) and the Australian Centre for Advanced
Computing and Communications (AC3).
This system is in use for the Australian contribution to the
production of simulated data for the Belle experiment, and for physics
analyses.
The Storage Resource Broker (SRB), from the San Diego Supercomputing
Centre, is used to provide a robust underlying data repository. A
federation of SRB servers has been established to share and manage
Belle data between the KEK laboratory, the mass data store at the
Australian National University (ANU) and satellite storage at each of
the compute clusters.
The globus toolkit is the underlying technology for the management of
the computing resources, and the despatching of jobs. A network aware
job scheduler has been developed. The scheduler queries the SRB
servers for location of data replicas, and arranges scheduling of processing and
production jobs on the compute resources according to a
static model of the network connectivity and dynamic assessment of the
relative system loads.