NanoCheckpoints: A Task-Based Asynchronous Dataflow Framework for Efficient and Scalable Checkpoint/Restart

Abstract

In this paper, we present NanoCheckpoints which is a lightweight software-based checkpoint/restart scheme for taskparallel HPC applications. We leverage OmpSs, a task-based OpenMP derivative programming model (PM) and its Nanos asynchronous dataflow runtime. NanoCheckpoints achieves minimal overheads by checkpointing only tasks' inputs which are available for free in the OmpSs PM. We evaluate NanoCheckpoints by both pure task-parallel shared memory benchmarks (up to 16 cores) and hybrid OmpSs+MPI applications (up to 1024 cores). The results indicate that NanoCheckpoints has on average overhead 3% for shared memory benchmarks. The dataflow semantics of Nanos, where both checkpointing and error recovery are asynchronous, allows NanoCheckpoints to scale at large core counts even when high error rates are present. For hybrid OmpSs+MPI benchmarks, NanoCheckpoints has very low overhead, on average 2%, and high scalability.

A taxonomy of task-based parallel programming technologies for high-performance computing

“…The process which runs a task is generally capable of detecting task faults. There are several examples of shared memory runtimes which are capable of detecting and correcting task faults within parallel regions [16,26].…”

Supporting automatic recovery in offloaded distributed programming models through MPI-3 techniques

“…Prior to the incorporation of the OmpSs offload functionality, "smart" C/R was introduced in the Nanos++ runtime to provide efficient fault-tolerance capabilities by benefiting from the PM semantics (leveraging the task data dependencies) [19]. This protected from memory faults reported by the OS.…”

A Runtime Heuristic to Selectively Replicate Tasks for Application-Specific Reliability Targets

“…The work of Subasi et al [38] is based on programmer knowledge in order to achieve effective partial replication. The NanoCheckpoints [36] and the message logging protocol proposed by Martsinkevich et al [27] address fail-stop errors of task-parallel computations. SSD [37] is designed by using machine learning techniques to mitigate silent errors in HPC applications.…”