Grid-Brick Event Processing Framework in GEPS

Abstract

Experiments like ATLAS at LHC involve a scale of computing and data management that greatly exceeds the capability of existing systems, making it necessary to resort to Grid-based Parallel Event Processing Systems (GEPS). Traditional Grid systems concentrate the data in central data servers which have to be accessed by many nodes each time an analysis or processing job starts. These systems require very powerful central data servers and make little use of the distributed disk space that is available in commodity computers. The Grid-Brick system, which is described in this paper, follows a different approach. The data storage is split among all grid nodes having each one a piece of the whole information. Users submit queries and the system will distribute the tasks through all the nodes and retrieve the result, merging them together in the Job Submit Server. The main advantage of using this system is the huge scalability it provides, while its biggest disadvantage appears in the case of failure of one of the nodes. A workaround for this problem involves data replication or backup.