Portable Checkpointing and Recovery in Heterogeneous Environments

Current approaches for checkpointing and recovery assume system homogeneity, where checkpointing and recovery are both performed on the same processor architecture and operating system configuration. Sometimes it is desirable or necessary to recover the failed computation on a different processor architecture, with possibly different byte-ordering and data-alignment specifications. This implies that checkpointing and recovery must be portable. We provide portability by means of a universal checkpoint format that allows object codes to resume execution from a checkpointed state, allowing for fast execution of already compiled code, rather than interpreting or compiling on the fly. This paper describes the system support needed to implement portable checkpoints, and the shadow checkpoint algorithm to checkpoint and recover a sequential process. Experimental results on three different architecture-operating system combinations demonstrate the checkpointing overhead and the cost of recov..