Authors: Lu Yang; Zhi-Qiang Liu; Jian-Feng Yan
Addresses: School of Computer Science and Technology, Soochow University, Suzhou, Jiangsu, China ' Department of Management and Innovation Systems, University of Salerno, Italy ' School of Computer Science and Technology, Soochow University, Suzhou, Jiangsu, China
Abstract: In big data parallel processing, parallel defects, e.g., data race and deadlock, are common causes that affect reliability of programs. Uncertainty in parallel processing characterises parallel defects, for which fuzziness of time sequence analysis plays an important role. To improve the performance of big data processing, we propose a multithreaded time sequence analysis approach based on type-2 fuzzy logic and hidden Markov model in this paper. Firstly, we collect a sample set of training data by carrying out extensive experiments for the target multi-threaded program with given observations. Secondly, we establish a time sequence analysis model to describe the inner relationship between the observations and time sequence of the target multi-threaded program. Thirdly, using this model we estimate the probability of each state sequence in all the target defect positions, with which we estimate the probability of defects for the corresponding observation sequence. To prove the scalability in a big data environment, we also use our approach to analyse a real concurrency defect in real world large-scale multi-thread programs. Our experiment results show that the average deviation using type-2 fuzzy logic is less than one fourth of the average deviation using type-1 fuzzy logic.
Keywords: time sequence analysis; type-2 fuzzy logic; hidden Markov model; HMM; big data.
International Journal of High Performance Systems Architecture, 2018 Vol.8 No.1/2, pp.114 - 125
Received: 09 Nov 2017
Accepted: 15 May 2018
Published online: 01 Aug 2018 *