Datacenter Congestion Control Research Articles

FastCredit: Expediting credit-based congestion control in datacenters

Receiver-driven transports that leverage hop-by-hop credits have the merits of fast convergence, short queueing, and strong congestion avoidability, making them high-efficiency in managing congestion in current high-speed data centers. However, these transport protocols treat long and latency-sensitive short flows equally, increasing flow completion times and affecting user experiences. Although flow scheduling mechanisms have been studied extensively, they are hard to be deployed into hop-by-hop credit-based transports. The root cause is that most traditional flow scheduling mechanisms mainly work in the long queue containing flows of various sizes, while credit-based proactive transports maintain the extremely short bounded queue.Based on this observation, this paper presents FastCredit, making the first attempt to expedite short-flow scheduling in hop-by-hop credit-based congestion control. In FastCredit, we schedule credit transmission at both receivers and switches to indirectly perform data flow scheduling, and we develop an ECN-based credit rate control mechanism to mitigate the problems caused by credit loss. Compared to the state-of-the-art credit-based transport protocol, FastCredit reduces average flow completion time to 0.71x and greatly improves the short flow transmission latency to 0.45x in realistic workloads. Especially, FastCredit reduces average flow completion time to 0.72x under incast circumstances and 0.56x in many-to-one traffic mode. Moreover, FastCredit still maintains the advantages of the short queue and high throughput and essentially reduces credit loss by employing credit rate control.

Reinforcement Learning for Datacenter Congestion Control

We approach the task of network congestion control in datacenters using Reinforcement Learning (RL). Successful congestion control algorithms can dramatically improve latency and overall network throughput. Until today, no such learning-based algorithms have shown practical potential in this domain. Evidently, the most popular recent deployments rely on rule-based heuristics that are tested on a predetermined set of benchmarks. Consequently, these heuristics do not generalize well to newly-seen scenarios. Contrarily, we devise an RL-based algorithm with the aim of generalizing to different configurations of real-world datacenter networks. We overcome challenges such as partial-observability, non-stationarity, and multi-objectiveness. We further propose a policy gradient algorithm that leverages the analytical structure of the reward function to approximate its derivative and improve stability. We show that these challenges prevent standard RL algorithms from operating within this domain. Our experiments, conducted on a realistic simulator that emulates communication networks' behavior, show that our method exhibits improved performance concurrently on the multiple considered metrics compared to the popular algorithms deployed today in real datacenters. Our algorithm is being productized to replace heuristics in some of the largest datacenters in the world.

Reinforcement Learning for Datacenter Congestion Control

We approach the task of network congestion control in datacenters using Reinforcement Learning (RL). Successful congestion control algorithms can dramatically improve latency and overall network throughput. Until today, no such learning-based algorithms have shown practical potential in this domain. Evidently, the most popular recent deployments rely on rule-based heuristics that are tested on a predetermined set of benchmarks. Consequently, these heuristics do not generalize well to newly-seen scenarios. Contrarily, we devise an RL-based algorithm with the aim of generalizing to different configurations of real-world datacenter networks. We overcome challenges such as partial-observability, nonstationarity, and multi-objectiveness. We further propose a policy gradient algorithm that leverages the analytical structure of the reward function to approximate its derivative and improve stability. We show that this scheme outperforms alternative popular RL approaches, and generalizes to scenarios that were not seen during training. Our experiments, conducted on a realistic simulator that emulates communication networks' behavior, exhibit improved performance concurrently on the multiple considered metrics compared to the popular algorithms deployed today in real datacenters. Our algorithm is being productized to replace heuristics in some of the largest datacenters in the world.

Harmonia: Explicit Congestion Notification and Credit-Reservation Transport Converged Congestion Control in Datacenters

Harmonia: Explicit Congestion Notification and Credit-Reservation Transport Converged Congestion Control in Datacenters

Comment on "Datacenter Congestion Control: Identifying what is essential and making it practical" by Aisha Mushtaq, et al, CCR, July 2019

We dispute the authors' claim that SRPT is the crucial factor in achieving good FCT performance in datacenter networks.

Datacenter congestion control

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.

DX: Latency-Based Congestion Control for Datacenters

Since the advent of datacenter networking, achieving low latency within the network has been a primary goal. Many congestion control schemes have been proposed in recent years to meet the datacenters’ unique performance requirement. The nature of congestion feedback largely governs the behavior of congestion control. In datacenter networks, where round trip times are in hundreds of microseconds, accurate feedback is crucial to achieve both high utilization and low queueing delay. Proposals for datacenter congestion control predominantly leverage explicit congestion notification (ECN) or even explicit in-network feedback to minimize the queuing delay. In this paper, we explore latency-based feedback as an alternative and show its advantages over ECN. Against the common belief that such implicit feedback is noisy and inaccurate, we demonstrate that latency-based implicit feedback is accurate enough to signal a single packet’s queuing delay in 10 Gb/s networks. Such high accuracy enables us to design a new congestion control algorithm, DX, that performs fine-grained control to adjust the congestion window just enough to achieve very low queuing delay while attaining full utilization. Our extensive evaluation shows that: 1) the latency measurement accurately reflects the one-way queuing delay in single packet level; 2) the latency feedback can be used to perform practical and fine-grained congestion control in high-speed datacenter networks; and 3) DX outperforms DCTCP with 5.33 times smaller median queueing delay at 1 Gb/s and 1.57 times at 10 Gb/s.

Selective Data Replication for Online Social Networks with Distributed Datacenters

Though the new OSN model, which deploys datacenters globally, helps reduce service latency, it causes higher inter-datacenter communication load. In Facebook, each datacenter has a full copy of all data, and the master datacenter updates all other datacenters, generating tremendous load in this new model. Distributed data storage, which only stores a user's data to his/her geographically closest datacenters mitigates the problem. However, frequent interactions between distant users lead to frequent inter-datacenter communication and hence long service latencies. In this paper, we aim to reduce inter-datacenter communications while still achieving low service latency. We first verify the benefits of the new model and present OSN typical properties that underlie the basis of our design. We then propose Selective Data replication mechanism in Distributed Datacenters ( $SD^3$ ). Since replicas need inter-datacenter data updates, datacenters in $SD^3$ jointly consider update rates and visit rates to select user data for replication; furthermore, $SD^3$ atomizes users’ different types of data (e.g., status update, friend post, music) for replication, ensuring that a replica always reduces inter-datacenter communication. $SD^3$ also incorporates three strategies to further enhance its performance: locality-aware multicast update tree, replica deactivation, and datacenter congestion control. The results of trace-driven experiments on the real-world PlanetLab testbed demonstrate the higher efficiency and effectiveness of $SD^3$ in comparison to other replication methods and the effectiveness of its three schemes.

DC-Vegas: A delay-based TCP congestion control algorithm for datacenter applications

TCP congestion control in datacenter networks is very different from in traditional network environments. Datacenter applications require TCP to provide soft real-time latency and have the ability to avoid incast throughput collapses. To meet the special requirements of datacenter congestion control, numerous solutions have been proposed, such as DCTCP, D2TCP, and D3. However, several deployment drawbacks, including significant modifications to switch hardware, the Operating System protocol stack, and/or upper-layer applications, as well as switch ECN requirements, which are not always available in already existing datacenters, limit deployment of these solutions. To address these deployment problems, in this paper, we proposed a delay-based TCP algorithm for datacenter congestion control, namely DC-Vegas. DC-Vegas combines the performance advantages of DCTCP with the deployment advantages of delay-based TCP Vegas. DC-Vegas can meet both soft real-time and incast avoidance requirements of datacenters, requiring minimal deployment modification to existing datacenter hardware/software (with only sender-side update and without ECN requirements). Experimental results obtained using the real datacenter test bed and an ns-2 simulator demonstrate that DC-Vegas has similar performance with the state-of-the-art Data Center TCP algorithm.

Letter to the Editor—A Proof of the Optimality of the Shortest Remaining Processing Time Discipline

Letter to the Editor—A Proof of the Optimality of the Shortest Remaining Processing Time Discipline