Abstract: A system and method for collecting dump information in a system having a plurality of computation processors to execute a parallel processing program, which minimizes a down time of the system due to a dump collection. The plurality of computation processors are interconnected via a network for executing the parallel processing program. When one of the computation processors has failed, abort operation means aborts the computation processors. Dump writing means writes dumps of the computation processors excluding the failed computation processor into an internal storage unit of the failed computation processor. After the dump writing means has finished writing the dumps, restarting means restarts the computation processors excluding the failed computation processor. Reading means transfers the dumps written in the failed computation processor and a dump of the failed computation processor to an external storage unit.