Datacenter congestion control

Abstract

Recent years have seen a slew of papers on datacenter congestion control mechanisms. In this editorial, we ask whether the bulk of this research is needed for the common case where congestion control involves hosts responding to simple congestion signals from the network and the performance goal is reducing some average measure of flow completion time. We raise this question because we find that, out of all the possible variations one could make in congestion control algorithms, the most essential feature is the switch scheduling algorithm. More specifically, we find that congestion control mechanisms that use Shortest-Remaining-Processing-Time (SRPT) achieve superior performance as long as the rate-setting algorithm at the host is reasonable. We further find that while SRPT's performance is quite robust to host behaviors, the performance of schemes that use scheduling algorithms like FIFO or Fair Queuing depend far more crucially on the rate-setting algorithm, and their performance is typically worse than what can be achieved with SRPT. Given these findings, we then ask whether it is practical to realize SRPT in switches without requiring custom hardware. We observe that approximate and deployable SRPT (ADS) designs exist, which leverage the small number of priority queues supported in almost all commodity switches, and require only software changes in the host and the switches. Our evaluations with one very simple ADS design shows that it can achieve performance close to true SRPT and is significantly better than FIFO. Thus, the answer to our basic question - whether the bulk of recent research on datacenter congestion control algorithms is needed for the common case - is no.

WoS-CoMS: Work Stealing-Based Congestion Management Scheme for SDN Programmable Networks

“…While the controller can quickly detect congestion, it is harder to end it immediately [4]. To achieve this, other servers in the relevant domain are usually involved.…”

“…Concerning the task processing order for congested data centers, [4] show that this problem stems from the scheduling policy used on the one hand for packet transfer by switches, and on the other hand for request processing within servers. [22] previously addressed congestion control and task scheduling only at the switch level for better packet routing.…”

“…[22] focused on bandwidth usage and hop count as key criteria for good scheduling. To improve control on servers, [4] proposes to prioritize tasks with the shortest remaining execution time (Shortest Remaining-Processing-Time (SRPT)) when scheduling, as opposed to a FIFO (First In First Out) or Fair Queeing approach, and to progressively move tasks out of the queue of tasks to be processed. However, heavier tasks may never be executed, or executed very late.…”

Review of SDN-based load-balancing methods, issues, challenges, and roadmap

“…Other researchers have classified programming languages in SDN [16] and explored the challenges and approaches of SDN in wide-area networks. Jungmin Son et al [17] study focused on the use of SDN in cloud computing, specifically on power optimization, traffic engineering, network virtualization, and security in data centre networks.…”

Explicit feedback for congestion control in Science DMZ cyberinfrastructures based on programable data-plane switches.

“…Legacy AQM algorithms have been adapted and implemented on P4 switches with certain limitations [65,66,67,68,69,70]. P4 also enabled the fast development of custom AQM algorithms such as P4-SRPT [71], P4-ABC [72], SP-FIO [73], FDPA [74], P4QoS [75] and QoSTCP [76]. The above schemes benefit from the advantage of accurate measurements on the data plane, which allows for more timely control actions on the network.…”