LoCo: Localizing Congestion

Datacenter congestion control remains to be a hard problem. The challenge is that while congestion control is usually implemented at the end-hosts, congestion itself can happen at network switches. Thus, the ability of end-hosts to react to congestion is fundamentally limited by the timeliness and precision of congestion signals from the network. Unfortunately, despite decades of research, the community is still in quest of such timely and precise congestion signals. LoCo takes a new approach to resolving the congestion control problem: it localizes the congestion to egress queues of the end-hosts, precisely where congestion control is implemented, while bounding both the worst-case queuing at each switch and the network utilization. LoCo achieves this by exploiting the structure in datacenter network topologies to perform network-wide admission control for each and every packet in the network. LoCo is a clean-slate design that requires changes in network switches as well as end-hosts NICs. We evaluate LoCo using both an end-to-end hardware implementation and large-scale simulations; our results show that LoCo not only achieves the above mentioned theoretical guarantees, but also achieves better performance than several state-of-the-art congestion control protocols over standard datacenter workloads.

[1]  Nick McKeown,et al.  Rate control protocol (rcp): congestion control to make flows complete quickly , 2008 .

[2]  Robert N. M. Watson,et al.  Queues Don't Matter When You Can JUMP Them! , 2015, NSDI.

[3]  David A. Maltz,et al.  Data center TCP (DCTCP) , 2010, SIGCOMM 2010.

[4]  VL2: a scalable and flexible data center network , 2011, Commun. ACM.

[5]  Randy H. Katz,et al.  FastLane: making short flows shorter with agile drop notification , 2015, SoCC.

[6]  Devavrat Shah,et al.  Fastpass , 2014, SIGCOMM.

[7]  Mark Handley,et al.  Re-architecting datacenter networks and stacks for low latency and high performance , 2017, SIGCOMM.

[8]  Minlan Yu,et al.  HPCC: high precision congestion control , 2019, SIGCOMM.

[9]  Amin Vahdat,et al.  Sincronia: near-optimal network design for coflows , 2018, SIGCOMM.

[10]  Lakshminarayanan Subramanian,et al.  One more bit is enough , 2005, SIGCOMM '05.

[11]  David A. Maltz,et al.  Network traffic characteristics of data centers in the wild , 2010, IMC '10.

[12]  Nick McKeown,et al.  Programmable Packet Scheduling at Line Rate , 2016, SIGCOMM.

[13]  Brighten Godfrey,et al.  Finishing flows quickly with preemptive scheduling , 2012, CCRV.

[14]  John K. Ousterhout,et al.  Homa: a receiver-driven low-latency transport protocol using network priorities , 2018, SIGCOMM.

[15]  Chuang Lin,et al.  Catch the Whole Lot in an Action: Rapid Precise Packet Loss Notification in Data Center , 2014, NSDI.

[16]  Dongsu Han,et al.  Credit-Scheduled Delay-Bounded Congestion Control for Datacenters , 2017, SIGCOMM.

[17]  Ming Zhang,et al.  Congestion Control for Large-Scale RDMA Deployments , 2015, Comput. Commun. Rev..

[18]  Richard M. Karp,et al.  An optimal algorithm for on-line bipartite matching , 1990, STOC '90.

[19]  Amin Vahdat,et al.  A scalable, commodity data center network architecture , 2008, SIGCOMM '08.

[20]  W. Marsden I and J , 2012 .

[21]  Sangeetha Abdu Jyothi,et al.  Is advance knowledge of flow sizes a plausible assumption? , 2019, NSDI.

[22]  Mark Handley,et al.  Congestion control for high bandwidth-delay product networks , 2002, SIGCOMM '02.

[23]  Niv Buchbinder,et al.  Online Algorithms for Maximum Cardinality Matching with Edge Arrivals , 2017, ESA.

[24]  Amin Vahdat,et al.  TIMELY: RTT-based Congestion Control for the Datacenter , 2015, Comput. Commun. Rev..

[25]  Thomas E. Anderson,et al.  High speed switch scheduling for local area networks , 1992, ASPLOS V.

[26]  Hakim Weatherspoon,et al.  Globally Synchronized Time via Datacenter Networks , 2019, IEEE/ACM Transactions on Networking.

[27]  Nick McKeown,et al.  pFabric: minimal near-optimal datacenter transport , 2013, SIGCOMM.

[28]  Nick McKeown,et al.  The iSLIP scheduling algorithm for input-queued switches , 1999, TNET.

[29]  Gautam Kumar,et al.  pHost: distributed near-optimal datacenter transport over commodity network fabric , 2015, CoNEXT.

[30]  Leslie G. Valiant,et al.  A Scheme for Fast Parallel Communication , 1982, SIAM J. Comput..

[31]  Hakim Weatherspoon,et al.  SoNIC: Precise Realtime Software Access and Control of Wired Networks , 2013, NSDI.