RADIC-based Message Passing Fault Tolerance System
Marcela Castro-León, Dolores Isabel Rexachs, Emilio Luque · 2012
Abstract—We present an analysis design of how to incorpo-rate a transparent fault tolerance system at socket level for message passing applications. The novel design changes the default socket model avoiding being unexpectedly closed due to a remote node failure. Moreover, a pessimistic log-based rollback recovery protocol added to this level makes possible restarting and re-executing a failed parallel process until the point of failure independently of the rest of the processes. This paper explains and analyzes the design time decisions. We tested and assessed them executing a master-worker (M/W) and Single Program Multiple Data (SPMD) applications which follow different communication patterns. Promising results of robustness in interprocess communication were obtained. Keywords-Fault-tolerance; High-Availability; RADIC; message passing; socket.