TY - GEN
T1 - A Weighted ECMP Load Balancing Scheme for Data Centers Using P4 Switches
AU - Ye, Jin Li
AU - Chen, Chien
AU - Huang Chu, Yu
PY - 2018/11/27
Y1 - 2018/11/27
N2 - This paper proposes a weighted Equal-cost multipath (W-ECMP) scheme in datacenters using a Programmable Data plane. Nowadays, data center networks face ongoing challenges for higher performance and greater efficiency. Equal-cost multi-path (ECMP) routing, the most common load-balancing scheme used on data center networks, tries to balance the traffic without awareness of the network congestion. To overcome this problem, CONGA and HULA are proposed successively. CONGA has implemented customized ASIC as a part of the switch fabric to collect congestion information distributively. Nevertheless, ASICs solution shows a lack of scalability and flexibility. However, although HULA solves some problems of CONGA by using p4 switch, best path selection strategy in HULA only selects and records the best path, which makes it easy to congest the best path, and HULA uses probes that waste bandwidth and deteriorate the performance. To solve the problems of HULA, W-ECMP chooses a path with a weighted probability to avoid congesting a path quickly. W-ECMP encapsulates the path congestion information into the normal traffic similar to the concept of Inband Network Telemetry (INT) in P4, which increases updating speed as the network loading increases. From experimental results, W - ECMP performs better than HULA does in average flow completion time (FCT).
AB - This paper proposes a weighted Equal-cost multipath (W-ECMP) scheme in datacenters using a Programmable Data plane. Nowadays, data center networks face ongoing challenges for higher performance and greater efficiency. Equal-cost multi-path (ECMP) routing, the most common load-balancing scheme used on data center networks, tries to balance the traffic without awareness of the network congestion. To overcome this problem, CONGA and HULA are proposed successively. CONGA has implemented customized ASIC as a part of the switch fabric to collect congestion information distributively. Nevertheless, ASICs solution shows a lack of scalability and flexibility. However, although HULA solves some problems of CONGA by using p4 switch, best path selection strategy in HULA only selects and records the best path, which makes it easy to congest the best path, and HULA uses probes that waste bandwidth and deteriorate the performance. To solve the problems of HULA, W-ECMP chooses a path with a weighted probability to avoid congesting a path quickly. W-ECMP encapsulates the path congestion information into the normal traffic similar to the concept of Inband Network Telemetry (INT) in P4, which increases updating speed as the network loading increases. From experimental results, W - ECMP performs better than HULA does in average flow completion time (FCT).
KW - data center load balancing
KW - network congestion
KW - programmable data plane
KW - weighted-ecmp
UR - http://www.scopus.com/inward/record.url?scp=85060225014&partnerID=8YFLogxK
U2 - 10.1109/CloudNet.2018.8549549
DO - 10.1109/CloudNet.2018.8549549
M3 - Conference contribution
AN - SCOPUS:85060225014
T3 - Proceedings of the 2018 IEEE 7th International Conference on Cloud Networking, CloudNet 2018
BT - Proceedings of the 2018 IEEE 7th International Conference on Cloud Networking, CloudNet 2018
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 22 October 2018 through 24 October 2018
ER -