Backoff Algorithms to Avoid TCP Incast in Data Center Networks

Shigeyuki Osada, Daichi Izumi, Shiden Kishimoto, Yukinobu Fukushima, Tokumi Yokohira

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

When a standard TCP implementation using the minimum retransmission timeout (RTOmin) of 200 ms is used in distributed file systems in data centers, a well-known throughput collapse called TCP Incast occurs in barrier synchronized applications, because 200 ms is too large as an RTOmin in data center networks. A TCP implementation (TCPFG) using a much smaller RTOmin attained by a fine-grained kernel timer is proposed in order to avoid TCP Incast. However, throughput still decreases in TCPFG when the number of servers in distributed file systems is large. In order to improve the performance of TCPFG, we have proposed a method (NOBTCP) which use a new TCP option which forces servers belonging to a barrier synchronized application to retransmit packets. Although the performance of the method is much better, it is practically hard to introduce the new TCP option. In this paper, in order to improve the performance of TCPFG, we propose using different backoff algorithms from the binary exponential backoff algorithm in TCPFG. Numerical examples show that TCPFG with each of the proposed backoff algorithms shows better performance than NOBTCP.

Original languageEnglish
Title of host publication9th International Conference on Information and Communication Technology Convergence
Subtitle of host publicationICT Convergence Powered by Smart Intelligence, ICTC 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages515-520
Number of pages6
ISBN (Electronic)9781538650400
DOIs
Publication statusPublished - Nov 16 2018
Event9th International Conference on Information and Communication Technology Convergence, ICTC 2018 - Jeju Island, Korea, Republic of
Duration: Oct 17 2018Oct 19 2018

Other

Other9th International Conference on Information and Communication Technology Convergence, ICTC 2018
CountryKorea, Republic of
CityJeju Island
Period10/17/1810/19/18

Fingerprint

Servers
Throughput
Data center
Kernel

Keywords

  • Data Center Networks
  • Incast
  • TCP
  • Throughput

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Computer Science Applications
  • Information Systems
  • Information Systems and Management
  • Artificial Intelligence

Cite this

Osada, S., Izumi, D., Kishimoto, S., Fukushima, Y., & Yokohira, T. (2018). Backoff Algorithms to Avoid TCP Incast in Data Center Networks. In 9th International Conference on Information and Communication Technology Convergence: ICT Convergence Powered by Smart Intelligence, ICTC 2018 (pp. 515-520). [8539630] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICTC.2018.8539630

Backoff Algorithms to Avoid TCP Incast in Data Center Networks. / Osada, Shigeyuki; Izumi, Daichi; Kishimoto, Shiden; Fukushima, Yukinobu; Yokohira, Tokumi.

9th International Conference on Information and Communication Technology Convergence: ICT Convergence Powered by Smart Intelligence, ICTC 2018. Institute of Electrical and Electronics Engineers Inc., 2018. p. 515-520 8539630.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Osada, S, Izumi, D, Kishimoto, S, Fukushima, Y & Yokohira, T 2018, Backoff Algorithms to Avoid TCP Incast in Data Center Networks. in 9th International Conference on Information and Communication Technology Convergence: ICT Convergence Powered by Smart Intelligence, ICTC 2018., 8539630, Institute of Electrical and Electronics Engineers Inc., pp. 515-520, 9th International Conference on Information and Communication Technology Convergence, ICTC 2018, Jeju Island, Korea, Republic of, 10/17/18. https://doi.org/10.1109/ICTC.2018.8539630
Osada S, Izumi D, Kishimoto S, Fukushima Y, Yokohira T. Backoff Algorithms to Avoid TCP Incast in Data Center Networks. In 9th International Conference on Information and Communication Technology Convergence: ICT Convergence Powered by Smart Intelligence, ICTC 2018. Institute of Electrical and Electronics Engineers Inc. 2018. p. 515-520. 8539630 https://doi.org/10.1109/ICTC.2018.8539630
Osada, Shigeyuki ; Izumi, Daichi ; Kishimoto, Shiden ; Fukushima, Yukinobu ; Yokohira, Tokumi. / Backoff Algorithms to Avoid TCP Incast in Data Center Networks. 9th International Conference on Information and Communication Technology Convergence: ICT Convergence Powered by Smart Intelligence, ICTC 2018. Institute of Electrical and Electronics Engineers Inc., 2018. pp. 515-520
@inproceedings{df53d16557494ecd890ccdde06a0585b,
title = "Backoff Algorithms to Avoid TCP Incast in Data Center Networks",
abstract = "When a standard TCP implementation using the minimum retransmission timeout (RTOmin) of 200 ms is used in distributed file systems in data centers, a well-known throughput collapse called TCP Incast occurs in barrier synchronized applications, because 200 ms is too large as an RTOmin in data center networks. A TCP implementation (TCPFG) using a much smaller RTOmin attained by a fine-grained kernel timer is proposed in order to avoid TCP Incast. However, throughput still decreases in TCPFG when the number of servers in distributed file systems is large. In order to improve the performance of TCPFG, we have proposed a method (NOBTCP) which use a new TCP option which forces servers belonging to a barrier synchronized application to retransmit packets. Although the performance of the method is much better, it is practically hard to introduce the new TCP option. In this paper, in order to improve the performance of TCPFG, we propose using different backoff algorithms from the binary exponential backoff algorithm in TCPFG. Numerical examples show that TCPFG with each of the proposed backoff algorithms shows better performance than NOBTCP.",
keywords = "Data Center Networks, Incast, TCP, Throughput",
author = "Shigeyuki Osada and Daichi Izumi and Shiden Kishimoto and Yukinobu Fukushima and Tokumi Yokohira",
year = "2018",
month = "11",
day = "16",
doi = "10.1109/ICTC.2018.8539630",
language = "English",
pages = "515--520",
booktitle = "9th International Conference on Information and Communication Technology Convergence",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - GEN

T1 - Backoff Algorithms to Avoid TCP Incast in Data Center Networks

AU - Osada, Shigeyuki

AU - Izumi, Daichi

AU - Kishimoto, Shiden

AU - Fukushima, Yukinobu

AU - Yokohira, Tokumi

PY - 2018/11/16

Y1 - 2018/11/16

N2 - When a standard TCP implementation using the minimum retransmission timeout (RTOmin) of 200 ms is used in distributed file systems in data centers, a well-known throughput collapse called TCP Incast occurs in barrier synchronized applications, because 200 ms is too large as an RTOmin in data center networks. A TCP implementation (TCPFG) using a much smaller RTOmin attained by a fine-grained kernel timer is proposed in order to avoid TCP Incast. However, throughput still decreases in TCPFG when the number of servers in distributed file systems is large. In order to improve the performance of TCPFG, we have proposed a method (NOBTCP) which use a new TCP option which forces servers belonging to a barrier synchronized application to retransmit packets. Although the performance of the method is much better, it is practically hard to introduce the new TCP option. In this paper, in order to improve the performance of TCPFG, we propose using different backoff algorithms from the binary exponential backoff algorithm in TCPFG. Numerical examples show that TCPFG with each of the proposed backoff algorithms shows better performance than NOBTCP.

AB - When a standard TCP implementation using the minimum retransmission timeout (RTOmin) of 200 ms is used in distributed file systems in data centers, a well-known throughput collapse called TCP Incast occurs in barrier synchronized applications, because 200 ms is too large as an RTOmin in data center networks. A TCP implementation (TCPFG) using a much smaller RTOmin attained by a fine-grained kernel timer is proposed in order to avoid TCP Incast. However, throughput still decreases in TCPFG when the number of servers in distributed file systems is large. In order to improve the performance of TCPFG, we have proposed a method (NOBTCP) which use a new TCP option which forces servers belonging to a barrier synchronized application to retransmit packets. Although the performance of the method is much better, it is practically hard to introduce the new TCP option. In this paper, in order to improve the performance of TCPFG, we propose using different backoff algorithms from the binary exponential backoff algorithm in TCPFG. Numerical examples show that TCPFG with each of the proposed backoff algorithms shows better performance than NOBTCP.

KW - Data Center Networks

KW - Incast

KW - TCP

KW - Throughput

UR - http://www.scopus.com/inward/record.url?scp=85059459276&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85059459276&partnerID=8YFLogxK

U2 - 10.1109/ICTC.2018.8539630

DO - 10.1109/ICTC.2018.8539630

M3 - Conference contribution

SP - 515

EP - 520

BT - 9th International Conference on Information and Communication Technology Convergence

PB - Institute of Electrical and Electronics Engineers Inc.

ER -