A Transparent, Incremental, Concurrent Checkpoint Mechanism for Real-time and Interactive Applications *

Abstract

TLB miss-based incremental, concurrent checkpoint mechanism for real-time and interactive applications called TIC-CKPT has been proposed, implemented and evaluated in this paper. TIC-CKPT allows setting the checkpoints overlaps with the execution of the chekcpointed processes. By resorting to tracking TLB misses to stop the first accesses to the target memory pages while saving memory address space to non-volatile storage. Meanwhile, a thread, which works in the privileged mode, copies the target pages to the designated memory buffer first, and then resumes the memory accesses. Finally the original pages in the designated memory buffer are used to construct a consistent state of the checkpointed process. From the experimental results, in contrast to a traditional concurrent checkpoint system, TIC-CKPT saves more than 2% of the checkpoint time and decreases the stopped time of the checkpointed process by around 10%. Moreover, concurrent incremental checkpointing has been designed and implemented in TIC-CKPT as well. Compared with a conventional incremental checkpoint approach, TIC-CKPT can reduce the downtime introduced by setting an incremental checkpoint to a great extent while the benchmarks keep the principle of locality.