--> Overview

Project Name: StarPlane
Granted on: June 22nd, 2005
Main applicant: Prof. dr. H.E. Bal, VU
Co-applicants: Dr. ir. C.TH.A.M. de Laat, UvA H.J. Bos, VU

GLANCE theme: Management and analysis

As witnessed by the rising popularity of overlay networks, applications increasingly demand more flexibility from their networks. In e-Science the need is especially pressing. As more and more sites collaborate via wide area networks (WANs) in e-Science experiments, latency and bandwidth become major issues. This makes the topology and dimensioning of the network of vital importance (e.g., because queuing in intermediate hops adds to the latency). Moreover, it would be desirable to allow for network partitioning to prevent cross-interference by applications (e.g., to shield TCP-friendly connections from newly developed aggressive protocols on high-speed WANs). Unfortunately, in existing networks the topology and dimensioning of networks is fixed, and the number of hops between two nodes is immutable. Similarly, it is impossible for applications to request partitioning of network resources.

Optical networking will change the way networks are used. New control technology permits an application, in principle, to set up one or more end-to-end "lightpaths", providing it with hundreds of Gbit/s of aggregate bandwidth fully dedicated to the application. This moves the bandwidth bottleneck, as the network outside the computer becomes much faster than the connections inside. Using lightpaths, applications may allocate networks much like they allocate physical memory. For instance, applications may configure and dimension their own network topology, which would then consist of true end-to-end lightwave connections without intermediate queuing, switching or cross-interference. This is in contrast to current work in overlay networks, where a virtual link often consists of multiple switching/routing nodes connected by different types of links.

Unfortunately, while it is technically feasible to set up lightpaths, there exist neither the management plane to let individual applications exploit this functionality directly, nor the knowledge of how to integrate optical networks with applications. The proposed StarPlane project addresses both of these concerns.