Abstract: A multi-node parallel processing system includes multiple nodes each including the capability to handle faults. When a fault is detected, a fault handling procedure is launched. This may include invoking a dump capture handler to stop execution of all applications as well as control the saving of dump information to slave dump databases. The node in which the fault occurred sends a broadcast message to all the other nodes to indicate the fault condition. In response to the message, each of the other nodes also captures predetermined dump information into slave dump databases. The system is then restarted. After the system has started up again, a master dump save routine is started on a master node. The master dump save routine then launches slave dump save routines in each of the nodes to coordinate the collection of dump information. The master dump save routine can query the information stored in each of the slave dump routines and select a subset of the information desired.