Exploit Every Cycle: Vectorized Time Series Algorithms on Modern Commodity CPUs

Conference paper

DOI: 10.1007/978-3-319-56111-0_2

Part of the Lecture Notes in Computer Science book series (LNCS, volume 10195)
Cite this paper as:
Tang B., Yiu M.L., Li Y., U L.H. (2017) Exploit Every Cycle: Vectorized Time Series Algorithms on Modern Commodity CPUs. In: Blanas S., Bordawekar R., Lahiri T., Levandoski J., Pavlo A. (eds) Data Management on New Hardware. IMDM 2016, ADMS 2016. Lecture Notes in Computer Science, vol 10195. Springer, Cham

Abstract

Many time series algorithms reduce the computation cost by pruning unpromising candidates with lower-bound distance functions. In this paper, we focus on an orthogonal research direction that further boosts the performance by unlocking the potentials of modern commodity CPUs. First, we conduct a performance profiling on existing algorithms to understand where does time go. Second, we design vectorized implementations for lower-bound and distance functions that can enjoy characteristics (e.g., data parallelism, caching, branch prediction) provided by CPU. Third, our vectorized methods are general and applicable to many time series problems such as subsequence search, motif discovery and kNN classification. Our experimental study on real datasets shows that our proposal can achieve up to 6 times of speedup.

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  • Bo Tang
    • 1
  • Man Lung Yiu
    • 1
  • Yuhong Li
    • 2
  • Leong Hou U
    • 2
  1. 1.Hong Kong Polytechnic UniversityHung HomHong Kong
  2. 2.University of MacauTaipaMacau

Personalised recommendations