Achieving high utilization with software-driven WAN

By C. Hong, Srikanth Kandula, Ratul Mahajan, Ming Zhang, Vijay Gill, M. Nanduri, Roger Wattenhofer. Conference on Applications, Technologies, Architectures, and Protocols for Computer Communication, 2013

1,287 citations recorded by the source corpus.

We present SWAN, a system that boosts the utilization of inter-datacenter networks by centrally controlling when and how much traffic each service sends and frequently re-configuring the network's data plane to match current traffic demand. But done simplistically, these re-configurations can also cause severe, transient congestion because different switches may apply updates at different times. We develop a novel technique that leverages a small amount of scratch capacity on links to apply updates in a provably congestion-free manner, without making any assumptions about the order and timing of updates at individual switches. Further, to scale to large networks in the face of limited forwarding table capacity, SWAN greedily selects a small set of entries that can best satisfy current demand. It updates this set without disrupting traffic by leveraging a small amount of scratch capacity in forwarding tables. Experiments using a testbed prototype and data-driven simulations of two production networks show that SWAN carries 60% more traffic than the current practice.

Subfields: Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/3cff8e39fb13d196cbcfed985bb185353e1861ae

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

Dynamic scheduling of network updates (2014, background): “Categories and Subject Descriptors: C.2.1 [ComputerCommunication Networks]: Network Architecture and Design— Network communications; C.2.3 [Computer-Communication Networks]: Network Operations—Network management Keywords: Software-defined networking; network update”
Packet-Level Telemetry in Large Datacenter Networks (2015, background): “We also plan to extend Everflow to the cloud provider’s wide area network [16, 19], where, similar to DCNs, servers and network traffic are under its control.”
Low Latency Geo-distributed Data Analytics (2015, background): “(2) Microsoft Bing Edge Dashboard: Microsoft’s Bing service maintains a running dashboard of its edge servers deployed worldwide.”
Pingmesh: A Large-Scale System for Data Center Network Latency Measurement and Analysis (2015, background): “Software deﬁned networking (SWAN [13], B4 [16]) are further introduced for better wide area network traﬃc engineering.”
Gaia: Geo-Distributed Machine Learning Approaching LAN Speeds (2017, background): “In practice, WAN bandwidth can fluctuate over time [35], and the WAN latency can be intolerably high for some ML algorithms.”
PCC Vivace: Online-Learning Congestion Control (2018, background): “Recent studies on SDN-based traffic engineering [17, 18] and network optimization for big-data systems [9] suggest a need for resource allocation at the transport layer.”
Advanced study of SDN/OpenFlow controllers (2013, background): “Understanding all benefits Google and Microsoft recently switched to SDN in their data centers [2, 3].”
Fastpass (2014, background): “Orchestra is an application-level mechanism; the actual transfers use TCP. SWAN [19] frequently reconﬁgures the network’s data plane to match demand from different services.”
A roadmap for traffic engineering in SDN-OpenFlow networks (2014, methodology): “SWAN [116] is a Software-driven WAN (SWAN) proposed by Microsoft, which utilizes policy rules to allow inter-data center WANs to carry significantly more traffic for higher-priority services, while maintaining fairness among similar services.”
CONGA: distributed congestion-aware load balancing for datacenters (2015, background): “The source leaf makes load balancing decisions based on per uplink congestion metrics, derived by taking the maximum of the local congestion at the uplink and the remote congestion for the path (or paths) to the destination leaf that originate at the uplink.”
Planck (2014, background): “B4 [20] and SWAN [15] monitor and shape trafﬁc at the edge of the network, but these projects operate on the order of seconds or minutes.”
HULA: Scalable Load Balancing Using Programmable Data Planes (2016, methodology): “Unlike prior work such as Hedera [8], SWAN [14], and B4 [15], which use a central controller to balance load every few minutes, CONGA is more responsive because it operates in the data plane, permitting it to make load-balancing decisions every few microseconds.”
FireFly (2014, neutral): “Other work minimizes congestion during updates [30].”
Distributed SDN Control: Survey, Taxonomy, and Challenges (2018, methodology): “4) B4 and SWAN: Google’s B4 [75] network leverages the logical centralization enabled by the SDN paradigm to deploy centralized TE in coexistence with the standard shortest-path routing for the purpose of increasing the utilization of the inter-data-center links (near 100”