Borg: the next generation

Muhammad Tirmazi; Adam Barker; Nan Deng; Md E. Haque; Zhijing Gene Qin; Steven Hand; Mor Harchol-Balter; John Wilkes

doi:10.1145/3342195.3387517

Borg: the next generation

Muhammad Tirmazi, Adam Barker, Nan Deng, Md E. Haque, Zhijing Gene Qin, Steven Hand, Mor Harchol-Balter, John Wilkes

School of Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

This paper analyzes a newly-published trace that covers 8 different Borg [35] clusters for the month of May 2019. The trace enables researchers to explore how scheduling works in large-scale production compute clusters. We highlight how Borg has evolved and perform a longitudinal comparison of the newly-published 2019 trace against the 2011 trace, which has been highly cited within the research community.

Our findings show that Borg features such as alloc sets are used for resource-heavy workloads; automatic vertical scaling is effective; job-dependencies account for much of the high failure rates reported by prior studies; the workload arrival rate has increased, as has the use of resource over-commitment; the workload mix has changed, jobs have migrated from the free tier into the best-effort batch tier; the workload exhibits an extremely heavy-tailed distribution where the top 1% of jobs consume over 99% of resources; and there is a great deal of variation between different clusters.

Original language	English
Title of host publication	Proceedings of the Fifteenth European Conference on Computer Systems (EuroSys '20)
Place of Publication	New York
Publisher	ACM
Pages	1-14
ISBN (Print)	9781450368827
DOIs	https://doi.org/10.1145/3342195.3387517
Publication status	Published - 15 Apr 2020
Event	Fifteenth European Conference on Computer Systems (EuroSys ’20) - Heraklion, Greece Duration: 27 Apr 2020 → 30 Apr 2020 Conference number: 15 https://www.eurosys2020.org/

Conference

Conference	Fifteenth European Conference on Computer Systems (EuroSys ’20)
Abbreviated title	EuroSys '20
Country/Territory	Greece
City	Heraklion
Period	27/04/20 → 30/04/20
Internet address	https://www.eurosys2020.org/

Keywords

Data centers
Cloud computing

Access to Document

10.1145/3342195.3387517

Tirmazi_Borg
Copyright © 2020 Owner/Author. This work has been made available online in accordance with publisher policies or with permission. Permission for further reuse of this content should be sought from the publisher or the rights holder. This is the final published version of the work, which was originally published at https://doi.org/10.1145/3342195.3387517
Final published version, 2.24 MB

Cite this

@inproceedings{6ccd0ffeea2044e7a9dd2090101148f6,

title = "Borg: the next generation",

abstract = "This paper analyzes a newly-published trace that covers 8 different Borg [35] clusters for the month of May 2019. The trace enables researchers to explore how scheduling works in large-scale production compute clusters. We highlight how Borg has evolved and perform a longitudinal comparison of the newly-published 2019 trace against the 2011 trace, which has been highly cited within the research community.Our findings show that Borg features such as alloc sets are used for resource-heavy workloads; automatic vertical scaling is effective; job-dependencies account for much of the high failure rates reported by prior studies; the workload arrival rate has increased, as has the use of resource over-commitment; the workload mix has changed, jobs have migrated from the free tier into the best-effort batch tier; the workload exhibits an extremely heavy-tailed distribution where the top 1% of jobs consume over 99% of resources; and there is a great deal of variation between different clusters.",

keywords = "Data centers, Cloud computing",

author = "Muhammad Tirmazi and Adam Barker and Nan Deng and Haque, {Md E.} and Qin, {Zhijing Gene} and Steven Hand and Mor Harchol-Balter and John Wilkes",

year = "2020",

month = apr,

day = "15",

doi = "10.1145/3342195.3387517",

language = "English",

isbn = "9781450368827",

pages = "1--14",

booktitle = "Proceedings of the Fifteenth European Conference on Computer Systems (EuroSys '20)",

publisher = "ACM",

address = "United States",

note = "Fifteenth European Conference on Computer Systems (EuroSys {\textquoteright}20), EuroSys '20 ; Conference date: 27-04-2020 Through 30-04-2020",

url = "https://www.eurosys2020.org/",

}

Tirmazi, M, Barker, A, Deng, N, Haque, ME, Qin, ZG, Hand, S, Harchol-Balter, M & Wilkes, J 2020, Borg: the next generation. in Proceedings of the Fifteenth European Conference on Computer Systems (EuroSys '20)., 30, ACM, New York, pp. 1-14, Fifteenth European Conference on Computer Systems (EuroSys ’20), Heraklion, Greece, 27/04/20. https://doi.org/10.1145/3342195.3387517

TY - GEN

T1 - Borg

T2 - Fifteenth European Conference on Computer Systems (EuroSys ’20)

AU - Tirmazi, Muhammad

AU - Barker, Adam

AU - Deng, Nan

AU - Haque, Md E.

AU - Qin, Zhijing Gene

AU - Hand, Steven

AU - Harchol-Balter, Mor

AU - Wilkes, John

N1 - Conference code: 15

PY - 2020/4/15

Y1 - 2020/4/15

N2 - This paper analyzes a newly-published trace that covers 8 different Borg [35] clusters for the month of May 2019. The trace enables researchers to explore how scheduling works in large-scale production compute clusters. We highlight how Borg has evolved and perform a longitudinal comparison of the newly-published 2019 trace against the 2011 trace, which has been highly cited within the research community.Our findings show that Borg features such as alloc sets are used for resource-heavy workloads; automatic vertical scaling is effective; job-dependencies account for much of the high failure rates reported by prior studies; the workload arrival rate has increased, as has the use of resource over-commitment; the workload mix has changed, jobs have migrated from the free tier into the best-effort batch tier; the workload exhibits an extremely heavy-tailed distribution where the top 1% of jobs consume over 99% of resources; and there is a great deal of variation between different clusters.

AB - This paper analyzes a newly-published trace that covers 8 different Borg [35] clusters for the month of May 2019. The trace enables researchers to explore how scheduling works in large-scale production compute clusters. We highlight how Borg has evolved and perform a longitudinal comparison of the newly-published 2019 trace against the 2011 trace, which has been highly cited within the research community.Our findings show that Borg features such as alloc sets are used for resource-heavy workloads; automatic vertical scaling is effective; job-dependencies account for much of the high failure rates reported by prior studies; the workload arrival rate has increased, as has the use of resource over-commitment; the workload mix has changed, jobs have migrated from the free tier into the best-effort batch tier; the workload exhibits an extremely heavy-tailed distribution where the top 1% of jobs consume over 99% of resources; and there is a great deal of variation between different clusters.

KW - Data centers

KW - Cloud computing

U2 - 10.1145/3342195.3387517

DO - 10.1145/3342195.3387517

M3 - Conference contribution

SN - 9781450368827

SP - 1

EP - 14

BT - Proceedings of the Fifteenth European Conference on Computer Systems (EuroSys '20)

PB - ACM

CY - New York

Y2 - 27 April 2020 through 30 April 2020

ER -

Borg: the next generation

Abstract

Conference

Keywords

Access to Document

Fingerprint

Cite this