A faster checkpointing and recovery algorithm with a hierarchical storage approach

• Published in: Eighth International Conference on High Performance Computing in Asia Pacific Region : proceedings : 30 November - 3 December, Beijing, China pp. 5 pp. - 402
• Main Authors: Gao, Wen, Mingyu Chen, Nanya, T.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 2005
• Subjects: Algorithm design and analysis; Checkpointing; Clustering algorithms; Computers; Concurrent computing; Detectors; Fault detection; Fault tolerant systems; High performance computing; Operating systems
• ISBN: 9780769524863; 0769524869
• DOI: 10.1109/HPCASIA.2005.2

Fault tolerance is an inevitable part of cluster operating system. In Score cluster system, it provides coordinated checkpointing, rollback recovery mechanism and watch-dog timer detector for fault tolerance. In the checkpointing algorithm in Score, disk write is the bottleneck. To eliminate disk write overhead, this paper proposes a new diskless checkpointing and rollback recovery algorithm. Since the proposed algorithm does not need to calculate parity and write the checkpointing data into disk, it is analyzed to be a faster checkpointing algorithm than the original one. Based on comparison, the recovery time of the proposed algorithm is also less. However, the cluster can not tolerant multiple transient failure using this diskless checkpointing algorithm. To compensate this drawback, a hierarchical storage strategy is adopted. An experimental result shows that this diskless algorithm with a hierarchical storage approach is fast and effective.