Parallel Algorithm for Local-best-match Time Series Subsequence Similarity Search on the Intel MIC Architecture

The paper touches upon the problem of local-best-match time series subsequence similarity search. The problem assumes that a query sequence and a longer time series are given, and the task is to find all the subsequences whose distance from the query is the minimal among their neighboring subsequences and distance from the query is under specified threshold. The Dynamic Time Warping (DTW) is used as a distance metric, which currently is recognized as the best similarity measure for most time series applications. However, computation of DTW costs too much despite the existing sophisticated software approaches. Existing hardware approaches to DTW computation involve GPU and FPGA and pay no regard to the Intel Many Integrated Core architecture. The paper proposes a parallel algorithm for solving this problem using both CPU and the Intel Xeon Phi many-core coprocessor. The implementation is based on the OpenMP parallel programming technology and offload execution mode, where part of the code and data is transmitted to the coprocessor. The algorithm utilizes a queue of subsequences on the processor side, which are uploaded to the coprocessor for the DTW computations. The results of experiments confirm the effectiveness of the algorithm.