Optimal Structure of Computing-Nodes with Limited Number  
Author Kenichiro Naruse


Co-Author(s) Kazunori Iwata; Toshio Nakagawa


Abstract Tasks with random processing times are executed successively. Two types of checkpoints can be placed at the end of tasks. Tasks are executed by several computing-nodes. Two types of checkpoints are called compare-checkpoint (CCP) and compare-and-storecheckpoint (CSCP). CCP needs a small time compared with CSCP to store checkpoint data, and CSCP needs a long time compared with CCP to store checkpoint data. We consider duplex and majority structure models for high-speed processing with limited computing-node. To find an optimal structure, we adopt two types of checkpoint schemes for two structures model. The problem is that in which structure model is optimal, and in what places we set suitable checkpoints. We derive how many numbers of CCP checkpoints between CSCP to minimize the execution time, which structure model is optimal to minimize execution time. The mean execution times per one cluster are obtained and optimal numbers that minimize them are derived analytically and numerically. We show the figures of the relationship between reliability and optimal checkpoint number and the relationship between reliability and optimal execution times, and propose a method to derive the intersection points of execution times between two structures.


Keywords Reliability, Checkpoint times, Duplex system, Majority system, High-speed processing, Limited computingnode
    Article #:  RQD25-53
Proceedings of 25th ISSAT International Conference on Reliability & Quality in Design
August 1-3, 2019 - Las Vegas, NV, U.S.A.