Efficiency of Buffer Caching in Computing-Intensive Workloads

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

With the recent advent of the fourth industrial revolution, computing-intensive workloads such as big data and machine learning emerge every day. Even though the workloads are computation-intensive, there are file I/Os to access storage, and we need to improve the I/O performance by using buffer caching. This paper analyzes the efficiency of the buffer caching in the emerging computing-intensive workloads, and observes some peculiar I/O patterns, which degrades the performance of the buffer caching significantly. To relieve this problem, we present a new buffer caching scheme for improving the I/O performance of computing-intensive workloads. Simulation experiments with real-world traces show that the proposed buffer caching scheme improves the cache miss rate against the well-known LRU buffer caching policy by 35.2% on average and up to 81.6%.

Original languageEnglish
Title of host publicationProceedings - 2020 7th International Conference on Information Science and Control Engineering, ICISCE 2020
EditorsShaozi Li, Ying Dai, Jianwei Ma, Yun Cheng
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages548-552
Number of pages5
ISBN (Electronic)9781728164069
DOIs
StatePublished - Dec 2020
Event7th International Conference on Information Science and Control Engineering, ICISCE 2020 - Changsha, Hunan, China
Duration: 18 Dec 202020 Dec 2020

Publication series

NameProceedings - 2020 7th International Conference on Information Science and Control Engineering, ICISCE 2020

Conference

Conference7th International Conference on Information Science and Control Engineering, ICISCE 2020
Country/TerritoryChina
CityChangsha, Hunan
Period18/12/2020/12/20

Keywords

  • LRU
  • buffer caching
  • computing-intensive workload
  • file I/O
  • storage

Fingerprint

Dive into the research topics of 'Efficiency of Buffer Caching in Computing-Intensive Workloads'. Together they form a unique fingerprint.

Cite this