Research Assistants: J.-C. Gomez, R. Pasquini
Sponsors: ONR, ARO, NSF, DoE
Conch is a topology-based message passing system for fail-safe distributed computing on heterogeneous platforms. It supports a variety of user-defined topologies, including bus, star, ring, mesh, tree etc. The system is able to detect process failure during execution and make up-calls to an application. With appropriate application intervention, startup of new process(es), state restoration and execution resumption is supported. Conch trees are used to enable high-efficiency data-combining and reduction operations in EcliPSe simulations. Thus, the Conch substrate is a key component of the EcliPSe architecture and supports EcliPSe's fault-tolerance. The Conch system is connection-based, using TCP/IP for messaging, and simple signal-based threading to implement computation and communication subtasks. All message routing is accomplished by communication threads. The system currently supports computations on a heterogeneous mix of SPARC (SunOS 4.x, SunOS 5.x), Sequent Symmetry, Intel i860, Silicon Graphics IRIX, and IBM RS/6000 environments.