PARALLEL CHECKPOINTING FACILITY ON A METASYSTEM

摘要

This article describes experiences on incorporating uncoordinated checkpointing and recovery facilities in a Java-based metacomputing system. Our case study is SUMA, a metasystem for execution of Java bytecode, both sequential and parallel. We have incorporated an algorithm that implements a communication-induced checkpointing protocol into SUMA. This algorithm induces processes to take additional local (forced) checkpoints and to log all in-transit messages to ensure consistent global checkpoints. Preliminary results about the performance overhead produced by the implementation of this algorithm are presented.