Exploit every cycle: Vectorized time series algorithms on modern commodity CPUs

Bo Tang, Man Lung Yiu, Yuhong Li, Leong Hou U

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review

1 Citation (Scopus)

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.
Original languageEnglish
Title of host publicationData Management on New Hardware - 7th International Workshop on Accelerating Data Analysis and Data Management Systems Using Modern Processor and Storage Architectures, ADMS 2016 and 4th International Workshop on In-Memory Data Management and Analytics, IMDM 2016, Revised Selected Papers
PublisherSpringer Verlag
Pages18-39
Number of pages22
ISBN (Print)9783319561103
DOIs
Publication statusPublished - 1 Jan 2017
Event7th International Workshop on Accelerating Data Analysis and Data Management Systems Using Modern Processor and Storage Architectures, ADMS 2016 and 4th International Workshop on In-Memory Data Management and Analytics, IMDM 2016 - New Delhi, India
Duration: 1 Sept 20161 Sept 2016

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume10195 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference7th International Workshop on Accelerating Data Analysis and Data Management Systems Using Modern Processor and Storage Architectures, ADMS 2016 and 4th International Workshop on In-Memory Data Management and Analytics, IMDM 2016
Country/TerritoryIndia
CityNew Delhi
Period1/09/161/09/16

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Exploit every cycle: Vectorized time series algorithms on modern commodity CPUs'. Together they form a unique fingerprint.

Cite this