INTEGRATION OF DISTRIBUTED HETEROGENEOUS COMPUTING RESOURCES FOR THE MPD EXPERIMENT WITH DIRAC INTERWARE

摘要

Computing and storage resources are essential for efficient Monte Carlo generation. At JINR there are several types of computing resources: Tier-1 and Tier-2 grid clusters, "Govorun" supercomputer, JINR Cloud, and NICA cluster. There are also EOS disk and dCache tape storage systems. In order to use them, users have to be aware of many details and differences between resources and keep track of load on all of them. The DIRAC Interware was adopted, configured, and expanded to fulfill requirements of massive centralized Monte Carlo generation for the Multi-Purpose Detector (MPD). For a year, all the infrastructure was used via DIRAC to run successfully around 500 000 jobs with an average duration of 5 h each. The use of DIRAC allowed for unified data access, performance estimation, and accounting across all resources.