HG-ckpt: A tybrid granularity checkpoint frame
Li Hong Xu, Kai Lü, Xu Zhou · 2012
Checkpoint-restart techniques are very important fault-tolerance approaches for these systems. As we know, the overhead of checkpoint will influence the system performance seriously. Incremental checkpoint is a well researched technique to reduce the overhead of checkpoint. However, huge memory footprints of parallel applications place severe limitations on scalability of conventional incremental checkpointing techniques. Recently, the emerging Phase Change Memory technology is becoming available and can replace disks with the fast accesses and byte-address features. In this paper, we propose and implement a new incremental checkpoint frame named HG-ckpt, which uses both page-level and byte-level granularity to detect dirty data. Our experiment results show that HG-ckpt can reduce checkpoint file size up to 50% without other overhead when compared to conventional page-level incremental checkpoint on PCM.