Improving datacenter performance and robustness with multipath TCP

Raiciu, Costin; Barré, Sébastien; Pluntke, Christopher; Greenhalgh, Adam; Wischik, Damon; Handley, Mark

doi:10.1145/2018436.2018467

Cited by 465 publications

(311 citation statements)

References 6 publications

Supporting

Mentioning

304

Contrasting

Unclassified

Order By: Relevance

“…Further, scaling such centralized approaches to very large numbers of flows in large scale data centers presents challenges in terms of switch state requirements. MPTCP [26] is a transport layer solution that splits flows into subflows that take different paths through the network, and modulates the rates of subflows based on congestion. However, in direct networks, MPTCP requires many subflows to probe the different paths, making it impractical for short flows.…”

Section: Challengesmentioning

confidence: 99%

“…MPTCP is a recent host-based transport layer solution for traffic engineering [26]. MPTCP relies on splitting each flow into multiple subflows that take different paths through the network, and modulates the amount of data transmitted on each subflow based on congestion, thus improving the network utilization.…”

Section: Mptcp In Hyperx Networkmentioning

confidence: 99%

“…In this section, we evaluate how MPTCP benefits from Dahu non-shortest path routing, and the additional improvements achieved using Dahu dynamic load balancing. We extended the htsim packet level simulator [11] (used to evaluate MPTCP in [26]), to simulate a (L=3, S=10, T =20) HyperX network with 100 Mbps links. This network has 1000 switches, 20,000 hosts and an oversubscription ratio of 1:4.…”

Section: Mptcp In Hyperx Networkmentioning

confidence: 99%

“…MPTCP [26] proposes a host based approach for multipath load balancing, by splitting a flow into multiple subflows and modulating how much data is sent over different subflows based on congestion. However, as a transport protocol, it does not have control over the network paths taken by subflows.…”

Section: Related Workmentioning

confidence: 99%

“…These restrictions on topology and routing also mean that higher level adaptive protocols like MPTCP [26] are unable to take advantage of the full capacity of direct networks because all paths are not exposed to them through routing/forwarding tables.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Dahu: Commodity switches for direct connect data center networks

Radhakrishnan

Tewari

Kapoor

et al. 2013

Architectures for Networking and Communications Systems

View full text Add to dashboard Cite

Solving "Big Data" problems requires bridging massive quantities of compute, memory, and storage, which requires a very high bandwidth network. Recently proposed direct connect networks like HyperX [1] and Flattened Butterfly [20] offer large capacity through paths of varying lengths between servers, and are highly cost effective for common data center workloads. However data center deployments are constrained to multi-rooted tree topologies like Fat-tree [2] and VL2 [16] due to shortest path routing and the limitations of commodity data center switch silicon.In this work we present Dahu 1 , simple enhancements to commodity Ethernet switches to support direct connect networks in data centers. Dahu avoids congestion hot-spots by dynamically spreading traffic uniformly across links, and forwarding traffic over non-minimal paths where possible. By performing load balancing primarily using local information, Dahu can act more quickly than centralized approaches, and responds to failure gracefully. Our evaluation shows that Dahu delivers up to 500% improvement in throughput over ECMP in large scale HyperX networks with over 130,000 servers, and up to 50% higher throughput in an 8,192 server Fat-tree network.

show abstract

Section: Challengesmentioning

confidence: 99%

Section: Mptcp In Hyperx Networkmentioning

confidence: 99%

Section: Mptcp In Hyperx Networkmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Dahu: Commodity switches for direct connect data center networks

Radhakrishnan

Tewari

Kapoor

et al. 2013

Architectures for Networking and Communications Systems

View full text Add to dashboard Cite

show abstract

Economic feasibility of multipath protocols in mobile Internet of Things applications

Sonntag

Suomi

2013

Concurrency and Computation

View full text Add to dashboard Cite

Internet of Things envisions some services that require both end-device mobility and high availability or high bandwidth. However, the current environment with a cellular network and single-path protocols may not guarantee a sufficient service level. Thus, this study evaluates the economic feasibility of using multipath protocols for increased availability and bandwidth when contracting with several cellular operators. By using cost-efficiency analysis, a model is developed for comparing multipath protocols with their singlepath alternatives. The usage of the model is demonstrated with two potential mobile application scenarios that could benefit either from increased availability or bandwidth. Finland is used as an example market. The benefits of multipath protocols were quantified with an extensive set of handset-based measurements from Finnish cellular networks. The results of this study indicate that the initial costs of a multipath infrastructure remain relatively small, and the total costs rapidly approach the operational costs as the number or active time of smart devices increase. Therefore, this study indicates that multipath protocols are economically feasible for certain mobile Internet of Things applications, particularly those requiring high bandwidth. Thus, new service models emerging in the Internet could create a real demand for multipath protocols in the future.

show abstract

Congestion‐free routing strategy in software defined data center networks

Chen

et al. 2015

Concurrency and Computation

View full text Add to dashboard Cite

Large-scale online services and distributed execution engines (i.e., MapReduce and Dryad) generate large volumes of traffic in data center networks. As a consequence, significant congestion can occur in the data center network. To the best of our knowledge, most existing approaches either focus on local congestionaware mechanisms, which have only a poor ability to handle asymmetry or use explicit congestion notification packets, which are difficult to implement directly in switch hardware. These methods are insufficient to solve the congestion problem. In this paper, we focus on a congestion-free routing strategy, resorting to the global view of the data center network in a software-defined networking controller. Specifically, a timeslot allocation was first conducted for the coming packets, and then the corresponding routing paths were computed for each packet. In view of the efficiency, the timeslot allocation algorithm follows a heuristic pattern, and the path selection is modeled as a bin-packing problem. Simulation results showed that the congestion-free routing strategy proposed here performs well in throughput, queuing, and end-to-end round-trip time.To design a congestion-free routing strategy, a discrete time-slotted system where the timeslot length can range from hundreds of milliseconds to minutes is considered [10,11]. In each timeslot t , a number of flows generated by the applications arrive at the data center, and each packet of flows must be forwarded to the destination without causing any congestion in the data center. Hence, a congestionfree routing strategy is designed by taking advantage of an SDN controller to monitor the traffic and has a central network control [12,13]. Consequently, the key design is to assign timeslots and routing paths for each packet. As shown in Figure 1, the SDN controller contains three main functions: timeslot allocation, path selection, and SDN communication. Through communicating with switches, the SDN controller determined overall network situations as well as informed switches of the network modification. Once an endpoint (source) called for a transmission, the operating system sent this demand to the SDN controller. The source provided the controller with two pieces CONGESTION-FREE ROUTING STRATEGY 5737 Timeslot allocation designWe consider a discrete time-slotted system where time is slotted. In timeslot t , each endpoint can transmit or receive at most one Maximum Transmission Unit (MTU)-sized packet. Thus, when a large amount of flow arrives at the data center network, the first task is to allocate transfer time for them to avoid congestion. Here, timeslot allocation was conducted at the packet level.A congestion-free routing strategy design should ensure that traffic can only be routed in a congestion-free way and that it will satisfy both the input and output bandwidth constraints. Traditionally, today's data center networks are often organized into tiers, with tiers having various

show abstract

Improving datacenter performance and robustness with multipath TCP

Cited by 465 publications

References 6 publications

Dahu: Commodity switches for direct connect data center networks

Dahu: Commodity switches for direct connect data center networks

Economic feasibility of multipath protocols in mobile Internet of Things applications

Congestion‐free routing strategy in software defined data center networks

Contact Info

Product

Resources

About