Distributed heterogeneous computing infrastructure for NICA data processing
Seminars
Laboratory of Information Technologies
Joint Laboratory Seminar
Date and Time: Tuesday, 21 May 2024, at 3:00 PM
Venue: room 310, Meshcheryakov Laboratory of Information Technologies, online on Webinar
Seminar topic: “Distributed heterogeneous computing infrastructure for NICA data processing”
Speaker: Igor Pelevanyuk
Abstract:
One of the key components in the implementation of the NICA project, alongside the accelerators and detector setups BM@N, MPD, and SPD, is the computational infrastructure necessary for processing, analysis, storage, and transfer of large volumes of experimental data. Since 2019, a distributed heterogeneous computing infrastructure based on the DIRAC Interware has been developed at the MLIT. It includes Tier1 and Tier2 clusters, the “Govorun” supercomputer, the NICA cluster, the DDC cluster, clouds of participating countries of JINR, the UNAM cluster in Mexico, and the IMDT cluster in Mongolia. To improve the efficiency of operation, monitoring and performance analysis techniques for distributed heterogeneous computing resources have been developed and implemented. A fundamentally new approach to performance analysis of tasks has been developed, allowing the analysis of hundreds of thousands of jobs and identification of inefficiently operating resources. A methodology for modeling task behavior in a distributed heterogeneous environment has been developed. Its use allows forecasting the course of execution of large task batches.
The built infrastructure and tools are used to execute jobs of the MPD, BM@N, and SPD experiments. In total, more than 3 million jobs have been completed with an average execution time of 8 hours.