CANEs: Composable Active Network Elements

Quarterly Status Report

Period: December 11, 1997 - March 11, 1998


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

Accomplishments in the Quarter

We attended the DARPA Active Networks PIs meeting in Tucson. Calvert served as co-leader (with Larry Peterson) of the Node O/S and Network Architecture Working Group. That group will be revising the document entitled "Architectural Framework for Active Networks" (see below) to reflect the discussion at the PIs meeting. Zegura served as leader of the Composable Services and Applications Working Group. That group will be producing a document to reflect the discussion at the PIs meeting. Minutes from the Composable Services working group meeting are available here.

We developed a demonstration showing the effects of smart discard policies on MPEG video streams. This demonstration makes use of our earlier work on application-specific congestion control. Four congestion control mechanisms are used (IP packet discard, MPEG frame discard, priority and GOP discard). The demonstration includes both an adaptive sender and a non-adaptive sender. Two experimental topologies are used, a single active network node and two active network nodes. This demonstration dramatically illustrates that active networking can help applications under congestion. We showed the demo at the DARPA Net PIs meeting in Tucson.

We began work on the formal methods aspects of a composition approach called LIANE (Language Independent Active Network Environment). Composition in LIANE is achieved in two steps. First the user selects an "underlying program" that is executed on behalf of the user. The underlying program is offered by the node and has provable properties. A node may offer multiple different underlying programs. The users then select or provide a set of "injected programs," that can be used to customize the underlying program on a per-flow basis. We provide restrictions on the form of the injected programs so that properties can be proven of the composite service.

The LIANE approach represents a dual to the U. of Pennsylvania PLAN approach. In PLAN, the packet carries the glue that specifies the program structure and the node contains the composable elements. In LIANE, the node provides allowed program structures and the packet carries (or specifies) the composable elements. We believe LIANE offers some advantages in the flexibility of the composition and the provability of the properties of the composite service.

We ported our algorithms for active network caching to the ANTS environment. This represents a step in establishing a collaborative ABONE application based on information retrieval, as discussed with Guttag and Wetherall (MIT) during the November architecture working group meeting in Atlanta. We discussed progress on this application with Wetherall (MIT) during the DARPA PIs meeting; MIT has some complementary work on protocols for cache access.

We extended our simulator, AN-Sim, to include a rich set of random number distributions and queueing disciplines. We are working on including our congestion control algorithms in the simulator to support experiments with multicast congestion control.

Publications and Presentations


Administrative Issues

None in this quarter.

Plans for Next Quarter

We will be heavily involved in producing the two working group documents mentioned above. These are due on June 1, 1998.

We will continue development of LIANE and associated framework for proving properties of composed active services. We plan a submission on formal reasoning about active networks to the International Conference on Network Protocols (ICNP'98) in early May.

We will extend AN-Sim to include both congestion control and multicast capability. This will support experiments on multicast congestion control to add to the unicast experiments that we have already completed.

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. Simulations with AN-Sim will be added.

We will explore the application of active networking to a new area, namely wide-area server selection. This ties into related work by Zegura and Bhattacharjee using non-active methods to support selection of a good server.

We are prepared to work with DARPA on scheduling an Active Networks PIs meeting in Atlanta in July 1997. We have established a local contact person (Linda Williams) to coordinate administrative details with Jeanette Barker.

Ellen Zegura
Last modified: Thu Mar 26 08:24:11 PST 1998