Adaptive data management in the ARC Grid middleware

Abstract

Abstract. The Advanced Resource Connector (ARC) Grid middleware was designed almost 10 years ago, and has proven to be an attractive distributed computing solution and successful in adapting to new data management and storage technologies. However, with an ever-increasing user base and scale of resources to manage, along with the introduction of more advanced data transfer protocols, some limitations in the current architecture have become apparent. The simple first-in first-out approach to data transfer leads to bottlenecks in the system, as does the built-in assumption that all data is immediately available from remote data storage. We present an entirely new data management architecture for ARC which aims to alleviate these problems, by introducing a three-layer structure. The top layer accepts incoming requests for data transfer and directs them to the middle layer, which schedules individual transfers and negotiates with various intermediate catalog and storage systems until the physical file is ready to be transferred. The lower layer performs all operations which use large amounts of bandwidth, i.e. the physical data transfer. Using such a layered structure allows more efficient use of the available bandwidth as well as enabling late-binding of jobs to data transfer slots based on a priority system. Here we describe in full detail the design and implementation of the new system.

EMI Execution Service implementation in ARC

“…Communication part is handled by WS framework of ARC with job processing implemented as separate module known as Grid Manager. Data staging is done in another reusable module which provides balanced data download/upload functionality [18]. And credentials delegation is implemented as a dedicated library of C++ classes which provides bridge between communication and job control modules.…”