Protocol Membership in Dependable Distributed Communication Systems - A Question of Brittleness 2003-01-0108
This paper describes results from fault injection experiments using heavy ions in the time-triggered communication protocol for safety critical distributed systems (TTP/C, C1 implementation). The observed results show that arbitrary faults in one erroneous node could cause inconsistencies in the cluster and thus jeopardize correctly working nodes and the whole communication system. The described inconsistencies resulted from either asymmetric value faults or slightly out of specification timing faults. This system behavior can be partly explained by too strict constraints on the fault handling algorithms using the membership agreement protocol.