Credit-Scheduled Delay-Bounded Congestion Control for Datacenters

Cho, Inho; Jang, Keon; Han, Dongsu

doi:10.1145/3098822.3098840

Cited by 180 publications

(63 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We see that DCTCP takes 6 RTTs (480 µs) to converge to fair share whereas Slytherin converges in 4 RTTs (320 µs). Because Slytherin provides faster convergence, it effectively mitigates utilization and fairness issues in multibottleneck scenarios, as reported in prior studies [21].…”

Section: Convergence Timementioning

confidence: 77%

Slytherin: Dynamic, Network-Assisted Prioritization of Tail Packets in Datacenter Networks

Rezaei

Malekpourshahraki

Vamanan

2018

2018 27th International Conference on Computer Communication and Networks (ICCCN)

View full text Add to dashboard Cite

Datacenter applications demand both low latency and high throughput; while interactive applications (e.g., Web Search) demand low tail latency for their short messages due to their partition-aggregate software architecture, many dataintensive applications (e.g., Map-Reduce) require high throughput for long flows as they move vast amounts of data across the network. Recent proposals improve latency of short flows and throughput of long flows by addressing the shortcomings of existing packet scheduling and congestion control algorithms, respectively. We make the key observation that long tails in the Flow Completion Times (FCT) of short flows result from packets that suffer congestion at more than one switch along their paths in the network. Our proposal, Slytherin, specifically targets packets that suffered from congestion at multiple points and prioritizes them in the network. Slytherin leverages ECN mechanism which is widely used in existing datacenters to identify such tail packets and dynamically prioritizes them using existing priority queues. As compared to existing state-of-the-art packet scheduling proposals, Slytherin achieves 18.6% lower 99 th percentile flow completion times for short flows without any loss of throughput. Further, Slytherin drastically reduces 99 th percentile queue length in switches by a factor of about 2x on average.

show abstract

Section: Convergence Timementioning

confidence: 77%

Slytherin: Dynamic, Network-Assisted Prioritization of Tail Packets in Datacenter Networks

Rezaei

Malekpourshahraki

Vamanan

2018

2018 27th International Conference on Computer Communication and Networks (ICCCN)

View full text Add to dashboard Cite

show abstract

“…Despite these innovative ideas, because these schemes tackle the general case with arbitrarily changing number of flows which interact in arbitrary ways, the schemes rely on slow, iterative convergence to the appropriate sending rates. As discussed in Section 1, other schemes, including EyeQ [28], NumFabric [40] and ExpressPass [8], also rely on iterative convergence. Such convergence requires many round trips (e.g., 50 RTTs in TIMELY, 31 RTTs in NUMFabric, and 25-30 RTTs in EyeQ), as illustrated in Figure 1 for a sender whose initial sending rate is 100% of the line rate and the target rate is 50%.…”

Section: Challengesmentioning

confidence: 99%

“…(3) NUMFabric [40] achieves more flexible and faster bandwidth allocation than TCP but still employs iterative convergence (e.g., 31 RTTs). And, (4) while ExpressPass [8] and NDP [23] target general congestion via receiver-based congestion control, neither scheme isolates receiver congestion. ExpressPass employs BIC-TCP iterative convergence which takes 20 RTTs for a datacenter network (Section 5.1); ExpressPass shows results only for a simple network.…”

Section: Introductionmentioning

confidence: 99%

Dart: Divide and Specialize for Fast Response to Congestion in RDMA-Based Datacenter Networks

Xue

Chaudhry

Vamanan

et al. 2020

IEEE/ACM Trans. Networking

View full text Add to dashboard Cite

Though Remote Direct Memory Access (RDMA) promises to reduce datacenter network latencies significantly compared to TCP (e.g., 10x), end-to-end congestion control in the presence of incasts is a challenge. Targeting the full generality of the congestion problem, previous schemes rely on slow, iterative convergence to the appropriate sending rates (e.g., TIMELY takes 50 RTTs). Several papers have shown that even in oversubscribed datacenter networks most congestion occurs at the receiver. Accordingly, we propose a divide-and-specialize approach, called Dart, which isolates the common case of receiver congestion and further sub-divides the remaining in-network congestion into the simpler spatially-localized and the harder spatially-dispersed cases. For receiver congestion, we propose direct apportioning of sending rates (DASR) in which a receiver for n senders directs each sender to cut its rate by a factor of n, converging in only one RTT. For the spatially-localized case, Dart provides fast (under one RTT) response by adding novel switch hardware for in-order flow deflection (IOFD) because RDMA disallows packet reordering on which previous load balancing schemes rely. For the uncommon spatially-dispersed case, Dart falls back to DCQCN. Small-scale testbed measurements and at-scale simulations, respectively, show that Dart achieves 60% (2.5x) and 79% (4.8x) lower 99 th -percentile latency, and similar and 58% higher throughput than InfiniBand, and TIMELY and DCQCN.

show abstract

“…Receivers prioritize the flow with fewest remaining bytes when assigning tokens, achieving near optimal performance. ExpressPass [170] is another technique that uses receiver-side credit packets to control sender-side rate; credit packets can be lost without consequence, and the loss rate is used to gauge the connection capacity. Homa [171] is a new connectionless protocol inspired by pHost and ExpressPass, which can significantly reduce latency with no network support.…”

Section: G Datacenter Network and The Incast Issuementioning

confidence: 99%

A Survey on Recent Advances in Transport Layer Protocols

Polese

Chiariotti

Bonetto

et al. 2019

IEEE Commun. Surv. Tutorials

116

View full text Add to dashboard Cite

Over the years, the Internet has been enriched with new available communication technologies, for both fixed and mobile networks and devices, exhibiting an impressive growth in terms of performance, with steadily increasing available data rates. The Internet research community has kept trying to evolve the transport layer protocols to match the capabilities of modern networks, in order to fully reap the benefits of the new communication technologies. This paper surveys the main novelties related to transport protocols that have been recently proposed, identifying three main research trends: (i) the evolution of congestion control algorithms, to target optimal performance in challenging scenarios, possibly with the application of machine learning techniques; (ii) the proposal of brand new transport protocols, alternative to the Transmission Control Protocol (TCP) and implemented in the user-space; and (iii) the introduction of multipath capabilities at the transport layer.

show abstract

Credit-Scheduled Delay-Bounded Congestion Control for Datacenters

Cited by 180 publications

References 27 publications

Slytherin: Dynamic, Network-Assisted Prioritization of Tail Packets in Datacenter Networks

Slytherin: Dynamic, Network-Assisted Prioritization of Tail Packets in Datacenter Networks

Dart: Divide and Specialize for Fast Response to Congestion in RDMA-Based Datacenter Networks

A Survey on Recent Advances in Transport Layer Protocols

Contact Info

Product

Resources

About