Implementation of Fault-Tolerant GridRPC Applications

In this paper, a task parallel application is implemented with Ninf-G which is a GridRPC system, and experimented on, using the Grid testbed in Asia Pacific, for three months. The application is programmed to run for a long time and typical fault patterns were gathered through tens of long executions. As a result, unstable network throughput was determined to be one of the biggest reasons for faults. Then, an important point for application developers is stressed, reminding them to avoid serious decline of task throughput during operations for faults, by timeout minimization for fault detection, background recovery and duplicate task assignments. This study also issues a steer for design of the automated fault-tolerant mechanism in a higher layer of the GridRPC framework.

• Type: Journal
• Year: 2006
• Where: GRID
• Authors: Yusuke Tanimura, Tsutomu Ikegami, Hidemoto Nakada, Yoshio Tanaka, Satoshi Sekiguchi
• Keywords: Distributed And Parallel Computing, GRID 2006, Long Executions, Task Parallel Application, Typical Fault Patterns