A Skeletal-Based Approach for the Development of Fault-Tolerant SPMD Applications

Abstract

International audienceDistributing applications over PC clusters to speed-up or size-up the execution is now commonplace. Yet efficiently tolerating faults of these systems is a major issue. To ease the addition of checkpoint-based fault tolerance at the application level, we introduce a Model for Low-Overhead Tolerance of Faults (MoLOToF) which is based on structuring applications using fault-tolerant skeletons. MoLOToF also encourages collaborations with the programmer and the execution environment. The skeletons are adapted to specific parallelization paradigms and yield what can be called fault-tolerant algorithmic skeletons. The application of MoLOToF to the SPMD parallelization paradigm results in our proposed FT-SPMD framework. Experiments show that the complexity for developing an application is small and the use of the framework has a small impact on performance. Comparisons with existing system-level checkpoint solutions, namely LAM/MPI and DMTCP, point out that FT-SPMD has a lower runtime overhead while being more robust when a higher level of fault tolerance is required

FT-GReLoSSS: A Skeletal-Based Approach towards Application Parallelization and Low-Overhead Fault Tolerance

“…Recovery from achieved checkpoints exhibited negligible overheads. These results confirm previous ones we achieved on more elementary but varied benchmark applications [1].…”

“…The first implementation is a Javaspace-based Master-Worker framework [14] which considers self-contained skeletons. The second implementation is a framework named FT-SPMD [1] which targets a broad family of SPMD applications named GReLoSSS (cf. Section IV).…”

“…FT-SPMD benefitted from several modifications in its architecture and API which resulted in a seriously improved version called FT-GReLoSSS (FTG for short) which is presented hereafter. Compared to [1] the framework architecture is presented more in-depth and it is validated using an industrial application (cf. Section IV).…”

“…A domain decomposition application such as Jacobi relaxations [17] will use that datastructure to store new results at the current superstep. A data circulation application such as in some parallel matrix multiplication algorithms [1] will use it to receive data for the next superstep from neighbouring processes.…”

“…Following a failure, application execution is interrupted and the most recent saved state is then used to resume execution. This paper presents FTG, a specialized framework derived from the MoLOToF programming model [1]. MoLOToF aims to facilitate the development of efficient parallel applications and their endowment with efficient application-level checkpointing.…”