Dahu: Commodity switches for direct connect data center networks

Radhakrishnan, Sivasankar; Tewari, Malveeka; Kapoor, Rishi; Porter, George; Vahdat, Amin

doi:10.1109/ancs.2013.6665176

Cited by 9 publications

(4 citation statements)

References 16 publications

(16 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Data-Center Traffic Engineering: Data-center networks have massive scale in terms of number of switches and hosts. Most traffic engineering work in data-center networks focuses on routing elephant flows as it is impractical to deal with all flows in a centralized manner [40,41,42,43,44,45]. To cope with the scalability problem, CONGA designs a distributed load balancing mechanism and implement it in switch hardware [46].…”

Section: Related Workmentioning

confidence: 99%

Optimizing Bulk Transfers with Software-Defined Optical WAN

Jin

Wei

et al. 2016

Proceedings of the 2016 ACM SIGCOMM Conference

115

View full text Add to dashboard Cite

Bulk transfer on the wide-area network (WAN) is a fundamental service to many globally-distributed applications. It is challenging to efficiently utilize expensive WAN bandwidth to achieve short transfer completion time and meet mission-critical deadlines. Advancements in software-defined networking (SDN) and optical hardware make it feasible and beneficial to quickly reconfigure optical devices in the optical layer, which brings a new opportunity for traffic management on the WAN. We present Owan, a novel traffic management system that optimizes wide-area bulk transfers with centralized joint control of the optical and network layers. Owan can dynamically change the network-layer topology by reconfiguring the optical devices. We develop efficient algorithms to jointly optimize optical circuit setup, routing and rate allocation, and dynamically adapt them to traffic demand changes. We have built a prototype of Owan with commodity optical and electrical hardware. Testbed experiments and large-scale simulations on two ISP topologies and one inter-DC topology show that Owan completes transfers up to 4.45× faster on average, and up to 1.36× more transfers meet their deadlines, as compared to prior methods that only control the network layer.

show abstract

Section: Related Workmentioning

confidence: 99%

Optimizing Bulk Transfers with Software-Defined Optical WAN

Jin

Wei

et al. 2016

Proceedings of the 2016 ACM SIGCOMM Conference

115

View full text Add to dashboard Cite

show abstract

“…Recent studies in the literature [15,38,40,43] propose other routing techniques for DCNs. As a matter of fact, the static flow-to-path mapping performed by ECMP does not take flow size and network utilization into account [65]. This may result in saturating commodity switch L3 buffers and degrading overall network performance [15].…”

Section: Routingmentioning

confidence: 99%

“…ECMP is the standard technique used for splitting traffic across equal-cost (uniform) paths. Nonetheless, it cannot fully utilize the available capacity in these multiple paths [65]. Non-uniform multiple paths, in turn, complicate the problem, as mechanisms must take more factors into account (such as path latency and current load).…”

Section: Research Challengesmentioning

confidence: 99%

Datacenter Networks and Relevant Standards

Fonseca

Boutaba

2015

Cloud Services, Networking, and Management

View full text Add to dashboard Cite

Datacenters are the core of cloud computing, and their network is an essential component to allow distributed applications to run efficiently and predictably [17]. However, not all datacenters provide cloud computing. In fact, there are two main types of datacenters: production and cloud. Production datacenters are often shared by one tenant or among multiple (possibly competing) groups, services and applications, but with low rate of arrival and departure. They run data-analytics jobs with relatively little variation in demands, and their size varies from hundreds of servers to tens of thousands of servers. Cloud datacenters, in contrast, have high rate of tenant arrival and departure (churn) [75], run both user-facing applications and inward computation, require elasticity (since application demands are highly variable), and consist of tens to hundreds of thousands of physical servers [12]. Moreover, clouds can be composed of several datacenters spread around the world. As an example, Google, Microsoft and Amazon (three of the biggest players in the market) have datacenters in four continents; and each company has over 900, 000 servers.This chapter presents a in-depth study of datacenter networks (DCNs), relevant standards and operation. Our goal here is three-fold: i) provide a detailed view of the net-Cloud Services, Networking and Management, First Edition.

show abstract

“…Where the notion of upstream/downstream switches is ambiguous[30], mechanisms like TTL expiry can also be leveraged to make sure HULA probes do not loop forever.…”

mentioning

confidence: 99%

Hula

Katta

Hira

Kim³

et al. 2016

Proceedings of the Symposium on SDN Research

277

View full text Add to dashboard Cite

Datacenter networks employ multi-rooted topologies (e.g., Leaf-Spine, Fat-Tree) to provide large bisection bandwidth. These topologies use a large degree of multipathing, and need a data-plane loadbalancing mechanism to effectively utilize their bisection bandwidth. The canonical load-balancing mechanism is equal-cost multipath routing (ECMP), which spreads traffic uniformly across multiple paths. Motivated by ECMP's shortcomings, congestion-aware load-balancing techniques such as CONGA have been developed. These techniques have two limitations. First, because switch memory is limited, they can only maintain a small amount of congestiontracking state at the edge switches, and do not scale to large topologies. Second, because they are implemented in custom hardware, they cannot be modified in the field. This paper presents HULA, a data-plane load-balancing algorithm that overcomes both limitations. First, instead of having the leaf switches track congestion on all paths to a destination, each HULA switch tracks congestion for the best path to a destination through a neighboring switch. Second, we design HULA for emerging programmable switches and program it in P4 to demonstrate that HULA could be run on such programmable chipsets, without requiring custom hardware. We evaluate HULA extensively in simulation, showing that it outperforms a scalable extension to CONGA in average flow completion time (1.6× at 50% load, 3× at 90% load).

show abstract

Dahu: Commodity switches for direct connect data center networks

Cited by 9 publications

References 16 publications

Optimizing Bulk Transfers with Software-Defined Optical WAN

Optimizing Bulk Transfers with Software-Defined Optical WAN

Datacenter Networks and Relevant Standards

Hula

Contact Info

Product

Resources

About