TCP incast avoidance based on connection serialization in data center networks

Shigeyuki Osada, Kazutoshi Kajita, Yukinobu Fukushima, Tokumi Yokohira

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

3 Citations (Scopus)

Abstract

In distributed file systems, a well-known congestion collapse called TCP Incast occurs because many servers send data to the same client and then many packets overflow the port buffer of the client link. Incast leads to throughput degradation in the network. In this paper, we propose two methods to avoid Incast based on the fact that only one connection can almost fully use the network bandwidth because the bandwidth-delay product is not so large in current data center networks. The first method is a method which completely serializes connection establishments. By the serialization, the number of packets in the port buffer becomes very small, which leads to Incast avoidance. The second method is a method which overlaps the slow start period of the next connection to the current established connection to improve throughput in the first method. Numerical results from extensive simulation runs show the effectiveness of our two proposed methods.

Original languageEnglish
Title of host publication2013 19th Asia-Pacific Conference on Communications, APCC 2013
PublisherIEEE Computer Society
Pages142-147
Number of pages6
ISBN (Print)9781467360500
DOIs
Publication statusPublished - 2013
Event2013 19th Asia-Pacific Conference on Communications, APCC 2013 - Denpasar, Indonesia
Duration: Aug 29 2013Aug 31 2013

Other

Other2013 19th Asia-Pacific Conference on Communications, APCC 2013
CountryIndonesia
CityDenpasar
Period8/29/138/31/13

Fingerprint

Throughput
Bandwidth
Servers
Degradation

ASJC Scopus subject areas

  • Computer Networks and Communications

Cite this

Osada, S., Kajita, K., Fukushima, Y., & Yokohira, T. (2013). TCP incast avoidance based on connection serialization in data center networks. In 2013 19th Asia-Pacific Conference on Communications, APCC 2013 (pp. 142-147). [6765931] IEEE Computer Society. https://doi.org/10.1109/APCC.2013.6765931

TCP incast avoidance based on connection serialization in data center networks. / Osada, Shigeyuki; Kajita, Kazutoshi; Fukushima, Yukinobu; Yokohira, Tokumi.

2013 19th Asia-Pacific Conference on Communications, APCC 2013. IEEE Computer Society, 2013. p. 142-147 6765931.

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

Osada, S, Kajita, K, Fukushima, Y & Yokohira, T 2013, TCP incast avoidance based on connection serialization in data center networks. in 2013 19th Asia-Pacific Conference on Communications, APCC 2013., 6765931, IEEE Computer Society, pp. 142-147, 2013 19th Asia-Pacific Conference on Communications, APCC 2013, Denpasar, Indonesia, 8/29/13. https://doi.org/10.1109/APCC.2013.6765931
Osada S, Kajita K, Fukushima Y, Yokohira T. TCP incast avoidance based on connection serialization in data center networks. In 2013 19th Asia-Pacific Conference on Communications, APCC 2013. IEEE Computer Society. 2013. p. 142-147. 6765931 https://doi.org/10.1109/APCC.2013.6765931
Osada, Shigeyuki ; Kajita, Kazutoshi ; Fukushima, Yukinobu ; Yokohira, Tokumi. / TCP incast avoidance based on connection serialization in data center networks. 2013 19th Asia-Pacific Conference on Communications, APCC 2013. IEEE Computer Society, 2013. pp. 142-147
@inproceedings{b543d8b7e39449aabf2bf786764d51b8,
title = "TCP incast avoidance based on connection serialization in data center networks",
abstract = "In distributed file systems, a well-known congestion collapse called TCP Incast occurs because many servers send data to the same client and then many packets overflow the port buffer of the client link. Incast leads to throughput degradation in the network. In this paper, we propose two methods to avoid Incast based on the fact that only one connection can almost fully use the network bandwidth because the bandwidth-delay product is not so large in current data center networks. The first method is a method which completely serializes connection establishments. By the serialization, the number of packets in the port buffer becomes very small, which leads to Incast avoidance. The second method is a method which overlaps the slow start period of the next connection to the current established connection to improve throughput in the first method. Numerical results from extensive simulation runs show the effectiveness of our two proposed methods.",
author = "Shigeyuki Osada and Kazutoshi Kajita and Yukinobu Fukushima and Tokumi Yokohira",
year = "2013",
doi = "10.1109/APCC.2013.6765931",
language = "English",
isbn = "9781467360500",
pages = "142--147",
booktitle = "2013 19th Asia-Pacific Conference on Communications, APCC 2013",
publisher = "IEEE Computer Society",
address = "United States",

}

TY - GEN

T1 - TCP incast avoidance based on connection serialization in data center networks

AU - Osada, Shigeyuki

AU - Kajita, Kazutoshi

AU - Fukushima, Yukinobu

AU - Yokohira, Tokumi

PY - 2013

Y1 - 2013

N2 - In distributed file systems, a well-known congestion collapse called TCP Incast occurs because many servers send data to the same client and then many packets overflow the port buffer of the client link. Incast leads to throughput degradation in the network. In this paper, we propose two methods to avoid Incast based on the fact that only one connection can almost fully use the network bandwidth because the bandwidth-delay product is not so large in current data center networks. The first method is a method which completely serializes connection establishments. By the serialization, the number of packets in the port buffer becomes very small, which leads to Incast avoidance. The second method is a method which overlaps the slow start period of the next connection to the current established connection to improve throughput in the first method. Numerical results from extensive simulation runs show the effectiveness of our two proposed methods.

AB - In distributed file systems, a well-known congestion collapse called TCP Incast occurs because many servers send data to the same client and then many packets overflow the port buffer of the client link. Incast leads to throughput degradation in the network. In this paper, we propose two methods to avoid Incast based on the fact that only one connection can almost fully use the network bandwidth because the bandwidth-delay product is not so large in current data center networks. The first method is a method which completely serializes connection establishments. By the serialization, the number of packets in the port buffer becomes very small, which leads to Incast avoidance. The second method is a method which overlaps the slow start period of the next connection to the current established connection to improve throughput in the first method. Numerical results from extensive simulation runs show the effectiveness of our two proposed methods.

UR - http://www.scopus.com/inward/record.url?scp=84902304009&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84902304009&partnerID=8YFLogxK

U2 - 10.1109/APCC.2013.6765931

DO - 10.1109/APCC.2013.6765931

M3 - Conference contribution

AN - SCOPUS:84902304009

SN - 9781467360500

SP - 142

EP - 147

BT - 2013 19th Asia-Pacific Conference on Communications, APCC 2013

PB - IEEE Computer Society

ER -