Network Fault Tolerance in Open MPI

High Performance Computing (HPC) systems are rapidly growing in size and complexity. As a result, transient and persistent network failures can occur on the time scale of application run times, reducing the productive utilization of these systems. The ubiquitous network protocol used to deal with such failures is TCP/IP, however, available implementations of this protocol provide unacceptable performance for HPC system users, and do not provide the high bandwidth, low latency communications of modern interconnects. This paper describes methods used to provide protection against several network errors such as dropped packets, corrupt packets, and loss of network interfaces while maintaining high-performance communications. Microbenchmark experiments using vendor supplied TCP/IP and O/S bypass low-level communications stacks over InfiniBand and Myrinet are used to demonstrate the high-performance characteristics of our protocol. The NAS Parallel Benchmarks are used to demonstrate the scalability and the minimal performance impact of this protocol. Communication level micro-benchmarks show that providing higher data reliability decreases bandwidth by up to 30% relative to unprotected communications, but provides performance improvements of a factor of four over TCP/IP running over InfiniBand DDR. In addition, application level benchmarks (communication/computation) show virtually no impact of the data reliability protocol on overall run-time.