This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
This project addresses the complex networking challenge presented by the emerging cloud computing model. Cloud providers must run a diverse set of client applications, each with potentially different networking demands, on shared data-center facilities. Traditionally, a datacenter network is configured to use the same routing process to choose the "best" route for each flow in a datacenter, regardless of the application. For example, Ethernet frequently performs shortest-path routing along a single spanning tree. Yet data center networks typically exhibit significant redundancy; routing along a single tree leaves many paths unused, sacrificing potential gains in reliability, isolation and performance.
Topology switching moves beyond this one-size-fits-all approach providing an architecture for fine-grained multi-topology networking. It allows applications to create custom routing systems within a data center; they can configure multiple logical topologies that, together, are tailored to their reliability and performance requirements. From a cloud provider's perspective, a topology-switched network increases efficiency by multiplexing potentially hundreds of topologies across the same shared physical network. The PIs are designing a scalable topology-switched routing platform that facilitates the exploration of application interfaces, management challenges, novel routing strategies, and performance benefits of this approach.
Ultimately, the project aims to develop a flexible topology management primitive that improves administrators' ability to effectively manage extremely large datacenter deployments. The research is also analyzing the benefits and costs of multi-topology networking. Additional outcomes of this proposal include a public release of the topology switching platform, enabling academic and industrial feedback and adoption.