Channel reservation protocol for over-subscribed channels and destinations

Jiang, Nan; Becker, Daniel U.; Dally, William J.

doi:10.1145/2503210.2503213

Cited by 17 publications

(5 citation statements)

References 46 publications

(66 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The other main class of congestion management is based on reservation such as Speculative reservation protocol (SRP) [17], channel reservation protocol (CRP) [24], Small-Message SRP (SMSRP) and Last-Hop Reservation Protocol (LHRP) [18]. SRP proposed by Jiang, et al is designed to avoid hot-spot congestion, but the traffic schedule based on network status information is not always available.…”

Section: Related Workmentioning

confidence: 99%

“…Many congestion control mechanisms have been proposed to resolve network congestion. SRP [17], CRP [24] and SMSRP [18] share the similar strategy that reserving network resources for each flow transmission to avoid congestion, but their implementation is too complex. Other class of congestion control is mainly about adaptive routing algorithm.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

HARE: History-Aware Adaptive Routing Algorithm for Endpoint Congestion in Networks-on-Chip

Jin

Dong

et al. 2018

Int J Parallel Prog

View full text Add to dashboard Cite

Endpoint congestion is one of the most challenging issues when designing low latency and high bandwidth on-chip interconnection networks. Tree saturation and head-ofline blocking caused by the endpoint congestion seriously decrease system throughput and increases network latency, leading to overall performance degradation. Adaptive routing algorithms utilize dynamic network states to route packets around congestion areas and potentially mitigate network congestions, but still cannot deal with endpoint congestions. Existing adaptive routing algorithms mainly take the current route information into account, and rarely use the route information of past packets. In this paper, we explore the route information of past packets, and led to the following novel observations that the virtual channel (VC) allocations of prior packets can be collected as useful information, and the tree saturation can be isolated through better VC selection strategy based on the past route information. Based on this observation, a novel history-aware adaptive routing algorithm for endpoint congestion, HARE, is proposed to improve network performance. We implement HARE based on the state-of-the-art routing algorithm, Footprint, and conduct extensive simulation experiments to compare it with our algorithm. The evaluation results show that our design alleviate the impact of tree saturation consistently and achieve high throughput on both synthetic and trace-driven workloads.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

HARE: History-Aware Adaptive Routing Algorithm for Endpoint Congestion in Networks-on-Chip

Jin

Dong

et al. 2018

Int J Parallel Prog

View full text Add to dashboard Cite

show abstract

“…Alternative implementations may rely on congestion avoidance at the transport level (such as TCP congestion control) or on the use of proactive reservations to avoid saturating the network, such as SRP, 34 CRP, 35 SMSRP, and LHRP. Alternative implementations may rely on congestion avoidance at the transport level (such as TCP congestion control) or on the use of proactive reservations to avoid saturating the network, such as SRP, 34 CRP, 35 SMSRP, and LHRP.…”

Section: Related Workmentioning

confidence: 99%

“…Injection throttling might be handled using the original QCN implementation or any other ECN mechanism, by identifying CNMs generated by endpoints and not intercepting them in the network switches. Alternative implementations may rely on congestion avoidance at the transport level (such as TCP congestion control) or on the use of proactive reservations to avoid saturating the network, such as SRP, CRP, SMSRP, and LHRP …”

Section: Related Workmentioning

confidence: 99%

Non‐minimal adaptive routing based on explicit congestion notifications

Benito

Vallejo

Izu

et al. 2018

Concurrency and Computation

View full text Add to dashboard Cite

Low-diameter networks require non-minimal adaptive routing to deal with varying traffic characteristics and avoid pathological performance. Such routing is based on local estimations of network congestion, based on link-level flow control credits. Dragonfly networks based on the extensions of commodity Ethernet networks using OpenFlow have been proposed for large HPC deployments with low power consumption. However, this network technology does not implement credit-based flow control. This work explores a range of routing solutions based on exploiting explicit congestion notification messages (in particular, 802.1Qau) to adapt the number of packets using non-minimal paths. The design (denoted QCN-Switch) associates a probability value to each output port. This value is updated to reflect downstream congestion and used to statistically divert traffic away from congested areas when the load is uneven, as in the case of adversarial traffic. A feedback comparison variant is designed to separate the cases of uniform traffic at saturation and adversarial traffic at low loads. Evaluation results show that QCN-Switch is a competitive design for both the uniform traffic and adversarial traffic. Furthermore, it is able to react to changes in traffic conditions in 0.4 ms or less. A sensitivity analysis identifies the best configuration and shows its performance trade-offs.

show abstract

“…Thus, the gradual deployment of ExpressPass into DCNs is a visible task in the future. Nonetheless, deploying credit-based protocols [14,15] like ExpressPass in DCNs will bring many challenges to the fairness of bandwidth allocation, especially in multitenant DCNs [25,26] . We show in Section 2 that simply mixing ExpressPass with reactive protocols deployed in real DCNs will cause serious trouble.…”

Section: Introductionmentioning

confidence: 99%

CCRP: Converging Credit-Based and Reactive Protocols in Datacenters

Bai

Dong

et al. 2021

Int J Parallel Prog

View full text Add to dashboard Cite

As the link speed has grown steadily from 10 Gbps to 100 Gbps, highspeed data center networks (DCNs) require more efficient congestion management. Therefore, proactive transports, especially credit-based congestion control, nowadays have drawn much attention because of fast convergence, near-zero queueing and low latency. However, in real deployment scenarios, it is hard to guarantee one protocol to be deployed in every host at one time. Thus, when the credit-based protocols are deployed into DCNs incrementally, the network will convert to multiprotocol state and face the following fundamental challenges: (i) unfairness, (ii) non-convergence, and (iii) high buffer occupancy. In this paper, we propose a new protocol, called CCRP, aiming for converging credit-based and reactive protocols in data centers. Targeting the mostly deployed protocol, i.e. DCQCN based on explicit congestion notification (ECN), in DCNs, CCRP leverages the forward ECN to detect the network congestion in data queue and optimizes feedback control of the credit-based transports. Our experiment results show that this design can address the unfair link allocation and converge with reactive protocols rapidly. Furthermore, CCRP achieves high utilization and low buffer occupancy at the same time.

show abstract

Channel reservation protocol for over-subscribed channels and destinations

Cited by 17 publications

References 46 publications

HARE: History-Aware Adaptive Routing Algorithm for Endpoint Congestion in Networks-on-Chip

HARE: History-Aware Adaptive Routing Algorithm for Endpoint Congestion in Networks-on-Chip

Non‐minimal adaptive routing based on explicit congestion notifications

CCRP: Converging Credit-Based and Reactive Protocols in Datacenters

Contact Info

Product

Resources

About