Event Date/Time: Oct 25, 2004
Report as Spam

Description

Whether it is image visualization, IP telephony, high-volume file transfers, or large-scale sciences like climate modeling, high-energy physics, and bioinformatics, these and other bandwidth-hungry applications are forcing us to re-think our current network theories and implementations. Today's network transport protocols are incapable of providing the network characteristics needed by large-scale applications. The highly flexible sharing relationships among geographically distributed computers are fundamentally changing the way we think about application development.

Wide-area network infrastructure is evolving in such a way that our current network theories and implementations are being obsoleted. In the past and to some degree today, the wide-area infrastructure is primarily packet-switched with electronic signaling and switching. However, this infrastructure is rapidly changing, with high-speed optical routers interconnected within a circuit-switched optical core and with slower packet-switched networks in the periphery. The ubiquitous TCP protocol and its variants were originally designed and optimized for low speed transfers on shared networks where bandwidth is a critical and limited resource. To improve the transport performance on high bandwidth-delay product links, researchers have proposed many TCP variants like Scalable TCP, HighSpeed TCP, FAST TCP, etc. However, the current gigabits-per-second (peak) throughputs achieved using these variants are for short durations under heavily simulated conditions with manual parameter tuning. Large-scale applications require orders of magnitude more throughput to be available during the entire lifetime of the application! Recently the lambda-grid community has proposed a range of rate-based protocols like Reliable-Blast UDP (RBUDP) and SABUL which have performed better than TCP in general. However, it is not clear whether TCP-based protocols, non-TCP-based protocols, or a combination of both will be required for hybrid networks. How the protocol suite must evolve to the changing infrastructure and how network provisioning will be deployed in these hybrid networks will be the key networking issues over the next five to ten years.

The PATHNets workshop invites original, unpublished, full papers that address research issues and real world case studies on the topics of interest that includes, but not limited to, the following:

• Dynamic provisioning over a centrally-managed core vs. a distributed-managed core.

• Real-time circuit set-up and scheduling algorithms.

• Dynamic call admission and online reconfiguration techniques for hybrid networks

• Transport protocols for a circuit-switched networks: TCP vs non-TCP.

• Transport protocols for hybrid networks: TCP based vs non-TCP based, or a dynamic adaptation of both.

• Heterogeneous routing and control algorithms that enable transport protocols to co-exist with an optical core.

• Adaptation of transport protocols to operate over WANs spanning multiple medium types, e.g. optical and wireless, which can result in specific hops seeing higher error rates, and subsequently, higher packet losses.

• Design, architecture, and evaluation of provisioning systems.

• Design, architecture, and evaluation of transport protocols.

PATHNets'04 is held in conjunction with BroadNets 2004, taking place in San Jose, CA, October 25 2004

Venue

Types