RCC: Enabling Receiver-Driven RDMA Congestion Control With Congestion Divide-and-Conquer in Datacenter Networks
RCC: Enabling Receiver-Driven RDMA Congestion Control With Congestion Divide-and-Conquer in Datacenter Networks
复制标题
RCC：通过数据中心网络中的拥塞分治法实现接收器驱动的 RDMA 拥塞控制
DOI：
10.1109/tnet.2022.3185105
复制
发表时间：
2023-02
期刊：
影响因子：
--
通讯作者：
Tao Huang
中科院分区：
文献类型：
--
作者：
Jiao Zhang;Xiaolong Zhong;Zirui Wan;Yu Tian;Tian Pan;Tao Huang
数据中心应用程序的开发导致需要具有微秒级延迟的端到端通信。因此，RDMA在数据中心网络中变得越来越普遍，以减轻传统软件网络堆栈处理速度慢所造成的延迟。然而，现有的RDMA拥塞控制机制在同时实现高吞吐量和低延迟方面远非最佳，或者需要额外的网络内功能支持。在本文中，通过观察大多数拥塞发生在数据中心网络的最后一跳，我们提出了RCC，一种接收器驱动的RDMA网络快速拥塞控制机制，结合了显式分配和迭代窗口调整。首先，提出了一种网络拥塞判别方法，将拥塞分为最后一跳拥塞和网络内拥塞两种类型。然后，提出了一种显式窗口分配机制来解决最后一跳拥塞问题，使发送端能够在一次rtt中收敛到合适的发送速率。针对网络内拥塞问题，提出了一种基于pid的迭代延迟窗口调整方案，实现了快速收敛和接近于零的排队延迟。RCC不需要额外的网络内支持，并且对硬件实现很友好。在我们的评估中，RCC的总体平均FCT（流程完成时间）比Homa， ExpressPass， DCQCN， TIMELY和HPCC好4.79 %。
The development of datacenter applications leads to the need for end-to-end communication with microsecond latency. As a result, RDMA is becoming prevalent in datacenter networks to mitigate the latency caused by the slow processing speed of the traditional software network stack. However, existing RDMA congestion control mechanisms are either far from optimal in simultaneously achieving high throughput and low latency or in need of additional in-network function support. In this paper, by leveraging the observation that most congestion occurs at the last hop in datacenter networks, we propose RCC, a receiver-driven rapid congestion control mechanism for RDMA networks that combines explicit assignment and iterative window adjustment. Firstly, we propose a network congestion distinguish method to classify congestions into two types, last-hop congestion and in-network congestion. Then, an Explicit Window Assignment mechanism is proposed to solve the last-hop congestion, which enables senders to converge to a proper sending rate in one-RTT. For in-network congestion, a PID-based iterative delay-based window adjustment scheme is proposed to achieve fast convergence and near-zero queuing latency. RCC does not need additional in-network support and is friendly to hardware implementation. In our evaluation, the overall average FCT (Flow Completion Time) of RCC is $4{\sim }79\%$ better than Homa, ExpressPass, DCQCN, TIMELY, and HPCC.
登录
查看更多内容
DOI：
--
发表时间：
2021
期刊：
--
影响因子：
--
作者：
Shiyu Liu;Ahmad Ghalayini;Mohammad Alizadeh;B. Prabhakar;M. Rosenblum;Anirudh Sivaraman
通讯作者：
Shiyu Liu;Ahmad Ghalayini;Mohammad Alizadeh;B. Prabhakar;M. Rosenblum;Anirudh Sivaraman
DOI：
10.1145/3341302.3342065
发表时间：
2019-08
期刊：
Proceedings of the ACM Special Interest Group on Data Communication
影响因子：
--
作者：
Kun Qian;Wenxue Cheng;Tong Zhang;Fengyuan Ren
通讯作者：
Kun Qian;Wenxue Cheng;Tong Zhang;Fengyuan Ren
DOI：
10.1109/icnp52444.2021.9651938
发表时间：
2021-11
期刊：
2021 IEEE 29th International Conference on Network Protocols (ICNP)
影响因子：
--
作者：
Jiao Zhang;Jiaming Shi;Xiaolong Zhong;Zirui Wan;Yuxing Tian;Tian Pan;Tao Huang
通讯作者：
Jiao Zhang;Jiaming Shi;Xiaolong Zhong;Zirui Wan;Yuxing Tian;Tian Pan;Tao Huang
DOI：
10.1145/1644893.1644918
发表时间：
2009-11
期刊：
--
影响因子：
--
作者：
Srikanth Kandula;S. Sengupta;A. Greenberg;Parveen Patel;R. Chaiken
通讯作者：
Srikanth Kandula;S. Sengupta;A. Greenberg;Parveen Patel;R. Chaiken
DOI：
10.1145/2901318.2901337
发表时间：
2016-04
期刊：
Proceedings of the Eleventh European Conference on Computer Systems
影响因子：
--
作者：
Ana Klimovic;Christos Kozyrakis;Eno Thereska;Binu John;Sanjeev Kumar
通讯作者：
Ana Klimovic;Christos Kozyrakis;Eno Thereska;Binu John;Sanjeev Kumar