CANEs: Composable Active Network Elements

Quarterly Status Report

Period: September 11 - December 11, 1997


Georgia Tech Faculty: GTE Laboratories: Research Assistants: Undergraduates:

Accomplishments in the Quarter

The architecture of the active network determines the nature of the interfaces that provide control over functionality. A range of possible interfaces can be considered ranging from trivial selection of a service from a fixed set to specification of a program to be interpreted by a language interpreter at a node. It is reasonable to expect that the usual tradeoff between flexibility and performance will apply to this spectrum. Hence, we believe that active networks will support multiple interfaces, or environments, through which network node behavior can be controlled: some users will value "raw" performance over flexibility, while others will opt for flexibility---in order to realize some higher-level, end-to-end performance gain. We have outlined the components of an active network architecture that we believe offers a number of advantages. First, it is very general, and is consistent with most active networking research of which we are aware. Second, it offers a potential evolutionary path from the current Internet to a fully active network. Finally, it supports optimization and competition in a number of dimensions. This work is described in a paper submitted to the IEEE Network Magazine special issue on Active Networking.

We hosted a meeting in Atlanta on November 14, 1997, to discuss an architectural framework for active networks. Participants included Calvert, Zegura and Bhattacharjee from Georgia Tech; Wetherall and Guttag from MIT; Braden from ISI and Peterson from Arizona. Minutes from the meeting are available here. The group is currently working on an architecture document that reflects the consensus reached during the meeting. The architecture document will be made available as soon as it is ready.

In the area of interoperability, we have also installed ANTS on three machines at Georgia Tech.

We have partially completed the next version of the active networking simulator AN-Sim. The new version has increased flexibility to allow modules to be easily added that describe basic functions such as the node processing model. These extensions allow modeling of heterogeneous networks that contain both active and non-active nodes. The new version also allows other research groups to simulate their AN processing models.

We extended the set of results on active networking and congestion control to include different congestion generation schemes. We have extended the tests to different multi-router scenarios and multiple senders and receivers. In addition we have developed new metrics to evaluate the performance of the active and in-active schemes.

Publications and Presentations


Administrative Issues

Bill McKinnon has accepted a position as a part-time research scientist in the CANEs group. Bill recently finished his PhD at North Carolina State University. He will also be working in the Information Technology and Telecommunications Lab of Georgia Tech Research Institute (GTRI).

Plans for Next Quarter

We have scheduled a meeting with Gisli Hjalmtysson and Jennifer Rexford of AT&T Labs Research for December 15, 1997 in Atlanta, Georgia. The purpose of this meeting is to discuss a collaboration in the area of active network support for quality of service in streaming applications. Assuming agreement is reached on such a collaboration, we will begin that work in the next quarter.

We plan a journal submission on the use of active networking to enhance best-effort congestion control. Portions of this work have already appeared in the High Performance Networking conference (HPN'97). More extensive experiments were run in Fall 1997.

We are prepared to work with DARPA on scheduling an Active Networks PIs meeting in Atlanta in February 1997, as previously discussed. If this is to occur, planning should commence very soon.

Building on Bhattacharjee's work on control-on-demand, we plan to continue developing architecture and mechanisms that i) support formal reasoning about active nodes' behavior, individually and collectively; ii) provide a simple set of building blocks for resource management; iii) support fast-path implementation techniques.

One outcome of the November 14, 1997, meeting was discussion of a group project. We will coordinate with MIT to contribute to development, deployment and testing of a group project.

Ellen Zegura
Last modified: Wed Dec 17 07:52:40 PST 1997