DraLCD:一种新的数据中心流量工程方法
|
摘要
流量均衡是为了避免网络拥塞而作为流量工程中的路由优化目标提出来的,由于数据中心网络的流量特性,使得传统IP网络的流量工程方法不一定适合.为此,本文在SDN(Software Defined Network)的框架下,提出了一种基于链路关键度的自适应负载均衡流量工程方法:DraLCD(Dynamic Routing Algorithm based on Link Critical Degree).该方法通过对全局视图的网络管控,并充分利用了网络中存在的冗余路径,在完成细粒度流量均衡的同时,能够降低控制器的计算开销以及与交换机之间的通信开销,最终完成路由优化的目标.最后,基于DraLCD设计的原型系统,通过在Mininet仿真平台中部署并进行仿真实验,与现有的等开销多路径路由算法ECMP(Equal-Cost MultiPath)以及GFF(Global First Fit)路由算法相比较,能够明显地提升网络性能.
Traffic balancing is proposed as a routing optimal goal in traffic engineering in order to avoid the network congestion. Because of data center traffic characteristics, traffic engineering of traditional IP networks may be not suitable to the data centers. Thus,we present a dynamic routing algorithm: DralCD( Dynamic Routing Algorithm based on Link Critical Degree),which is based on the link critical degree within SDN( Software Defined Network) framework. By the global visibility and making full use of the redundant paths in the network,DraLCD can realize the fine-grained traffic balancing while reducing the computing cost of controller and communication overhead between controller and switches, and finally achieves the goal of routing optimal. Furthermore,we design and perform the prototype of DraLCD on Mininet platform, and compare it with two other popular algorithms: ECMP( Equal-Cost Multi-Path) and GFF( Global First Fit). In our evaluations, the results show that DraLCD can significantly improve the network performance compared with the other two algorithms.
Traffic balancing is proposed as a routing optimal goal in traffic engineering in order to avoid the network congestion. Because of data center traffic characteristics, traffic engineering of traditional IP networks may be not suitable to the data centers. Thus,we present a dynamic routing algorithm: DralCD( Dynamic Routing Algorithm based on Link Critical Degree),which is based on the link critical degree within SDN( Software Defined Network) framework. By the global visibility and making full use of the redundant paths in the network,DraLCD can realize the fine-grained traffic balancing while reducing the computing cost of controller and communication overhead between controller and switches, and finally achieves the goal of routing optimal. Furthermore,we design and perform the prototype of DraLCD on Mininet platform, and compare it with two other popular algorithms: ECMP( Equal-Cost Multi-Path) and GFF( Global First Fit). In our evaluations, the results show that DraLCD can significantly improve the network performance compared with the other two algorithms.
引文
[1]Awduche D,Chiu A,Elwalid A,et al.Overview and Principles of Internet Traffic Engineering[EB/OL].https://w w w.rfc-editor.org/rfc/rfc3272.txt,2002-05/2015-03.
[2]Theophilus B,Aditya A,David A M.Network traffic characteristics of datacenter in the w ild[A].IM C'10:Proceedings of the 10th ACM SIGCOM M Conference on Internet M easurement[C].New York:ACM,2010.267-280.
[3]Ho T V,Yves D,Olivier B,et al.Traffic engineering for multiple spanning tree protocol in large data centers[A].ITC'11:Proceedings of the 23rd International Teletraffic Congress[C].San Francisco:IEEE,2011.23-30.
[4]Albert G,James R H,Navendu J,et al.VL2:A scalable and flexible data center netw ork[J].Communications of the ACM,2011,54(3):95-104.
[5]Wu X,Yang X W.DARD:Distributed adaptive routing for datacenter netw orks[A].ICDCS’12:Proceedings of 32nd IEEE International Conference on Distributed Computing Systems[C].Piscataw ay:IEEE,2012.32-41.
[6]Ford A,Raiciu C,Handley M,et al.Architectural Guidelines for M ultipath TCP Development[EB/OL].https://w w w.rfc-editor.org/rfc/rfc6182.txt,2011-03/2015-03.
[7]Mohammad A,Tom E,Sarang D,et al.CONGA:Distributed congestion aw are load balancing for datacenters[A].SIGCOM M'14:Proceedings of the 2014 ACM conference on SIGCOM M[C].New York:ACM,2014.503-514.
[8]Theophilus B,Ashok A,Aditya A,et al.Micro TE:Fine grained traffic engineering for data centers[A].Co NEXT'11:Proceedings of the Seventh Conference on emerging Netw orking Experiments and Technologies[C].New York:ACM,2011.8-8.
[9]Ian F A,Lee A,Wang P,et al.A roadmap for traffic engineering in SDN-Open Flow netw orks[J].Computer Netw orks,2014,71(1):1-30.
[10]Mohammad Al,Sivasankar R,Barath R,et al.Hedera:Dynamic flow scheduling for data-center netw orks[A].NSDI'10:Proceedings of the 7th USENIX Conference on Netw orked Systems Design and Implementation[C].Berkeley:ACM,2010.19-19.
[11]Andrew R,Wonho K,Praveen Y.Mahout:Low-overhead datacenter traffic management using end-host-based elephant detection[A].INFOCOM’11 Proceedings IEEE INFOCOM 2011[C].Piscataw ay:IEEE,2011.1629-1637.
[12]Andrew R,Jeffrey C,Jean T,et al.Devo Flow:Scaling flow management for high-performance netw orks[A].SIGCOM M'11:Proceedings of the 2011 ACM Conference on SIGCOM M[C].New York:ACM,2011.254-265.
[13]Jonathan P,Amy O,Hari B,et al.Fastpass:A centralized zero-queue datacenter netw ork[A].SIGCOM M’14:Proceedings of the 2014 ACM Conference on SIGCOM M[C].Chicago:ACM,2014.307-318.
[14]Yukihiro N,Kazuki H,Chunghan Lee,et al.Domain Flow:Practical flow management method using multiple flow tables in commodity sw itches[A].Co NEXT'13:Proceedings of the Ninth ACM Conference on Emerging Netw orking Experiments And Technologies[C].New York:ACM,2013.399-404.
[15]Siddhartha S,David S,Sunghwan I,et al.Scalable optimal flow routing in datacenter via local link balancing[A].Co NEXT'13:Proceedings of the ninth ACM Conference on Emerging Netw orking Experiments and Technologies[C].New York:ACM,2014.151-162.
[16]Keqiang H,Eric R,Kanak A,et al.Presto:Edge-based load balancing for fast datacenter netw orks[A].SIGCOM M’15:Proceedings of the 2015 ACM Conference on SIGCOM M[C].New York:ACM,2015.465-478.
[2]Theophilus B,Aditya A,David A M.Network traffic characteristics of datacenter in the w ild[A].IM C'10:Proceedings of the 10th ACM SIGCOM M Conference on Internet M easurement[C].New York:ACM,2010.267-280.
[3]Ho T V,Yves D,Olivier B,et al.Traffic engineering for multiple spanning tree protocol in large data centers[A].ITC'11:Proceedings of the 23rd International Teletraffic Congress[C].San Francisco:IEEE,2011.23-30.
[4]Albert G,James R H,Navendu J,et al.VL2:A scalable and flexible data center netw ork[J].Communications of the ACM,2011,54(3):95-104.
[5]Wu X,Yang X W.DARD:Distributed adaptive routing for datacenter netw orks[A].ICDCS’12:Proceedings of 32nd IEEE International Conference on Distributed Computing Systems[C].Piscataw ay:IEEE,2012.32-41.
[6]Ford A,Raiciu C,Handley M,et al.Architectural Guidelines for M ultipath TCP Development[EB/OL].https://w w w.rfc-editor.org/rfc/rfc6182.txt,2011-03/2015-03.
[7]Mohammad A,Tom E,Sarang D,et al.CONGA:Distributed congestion aw are load balancing for datacenters[A].SIGCOM M'14:Proceedings of the 2014 ACM conference on SIGCOM M[C].New York:ACM,2014.503-514.
[8]Theophilus B,Ashok A,Aditya A,et al.Micro TE:Fine grained traffic engineering for data centers[A].Co NEXT'11:Proceedings of the Seventh Conference on emerging Netw orking Experiments and Technologies[C].New York:ACM,2011.8-8.
[9]Ian F A,Lee A,Wang P,et al.A roadmap for traffic engineering in SDN-Open Flow netw orks[J].Computer Netw orks,2014,71(1):1-30.
[10]Mohammad Al,Sivasankar R,Barath R,et al.Hedera:Dynamic flow scheduling for data-center netw orks[A].NSDI'10:Proceedings of the 7th USENIX Conference on Netw orked Systems Design and Implementation[C].Berkeley:ACM,2010.19-19.
[11]Andrew R,Wonho K,Praveen Y.Mahout:Low-overhead datacenter traffic management using end-host-based elephant detection[A].INFOCOM’11 Proceedings IEEE INFOCOM 2011[C].Piscataw ay:IEEE,2011.1629-1637.
[12]Andrew R,Jeffrey C,Jean T,et al.Devo Flow:Scaling flow management for high-performance netw orks[A].SIGCOM M'11:Proceedings of the 2011 ACM Conference on SIGCOM M[C].New York:ACM,2011.254-265.
[13]Jonathan P,Amy O,Hari B,et al.Fastpass:A centralized zero-queue datacenter netw ork[A].SIGCOM M’14:Proceedings of the 2014 ACM Conference on SIGCOM M[C].Chicago:ACM,2014.307-318.
[14]Yukihiro N,Kazuki H,Chunghan Lee,et al.Domain Flow:Practical flow management method using multiple flow tables in commodity sw itches[A].Co NEXT'13:Proceedings of the Ninth ACM Conference on Emerging Netw orking Experiments And Technologies[C].New York:ACM,2013.399-404.
[15]Siddhartha S,David S,Sunghwan I,et al.Scalable optimal flow routing in datacenter via local link balancing[A].Co NEXT'13:Proceedings of the ninth ACM Conference on Emerging Netw orking Experiments and Technologies[C].New York:ACM,2014.151-162.
[16]Keqiang H,Eric R,Kanak A,et al.Presto:Edge-based load balancing for fast datacenter netw orks[A].SIGCOM M’15:Proceedings of the 2015 ACM Conference on SIGCOM M[C].New York:ACM,2015.465-478.