CANEs: Composable Active Network Elements

Quarterly Status Report

Period: December 11, 1998 - March 11, 1999

http://www.cc.gatech.edu/projects/canes

Participants

Georgia Tech Faculty and Staff: University of Kentucky Faculty: GTE Laboratories: Research Assistants: PostDoc:

Accomplishments in the Quarter

We continued development of Odyssey, which includes an active Node operating system called Bowman and the CANEs execution environment. Bowman supports three primary abstractions: channels, a-flows, and a state store. A generic packet filtering mechanism demultiplexes packets from input channels to a-flows. A-flows are user threads within Bowman which perform processing on behalf of packets. The state store provides a tuple space for the sharing of information between a-flows. Bowman also provides an interface for the dynamic extension of the NodeOS. The CANEs environment is implemented as a Bowman extension, and CANEs underlying programs are run as a-flows within Bowman. Per-flow/per-channel output queues were added to Bowman, currently round-robin FIFO output scheduling is available. A heavy-weight timer mechanism was added to Bowman which allows timer handlers to run in the a-flow that set the timer. An Odyssey code server was developed and is being tested. This server will allow Odyssey nodes to request and share code with one another. Generic structures and an associated interface were created for unicast and multicast routing tables storage and use. A source based multicast routing protocol for multicasting between Odyssey nodes was implemented and is being tested.

We have been planning for near and far term demos with our Team-4 counterparts. We have had conversations with UCSC to define an interface to Odyssey for the addition and operation of routing protocols such as theirs. We also held a series of conference calls with UMASS and TASC to plan porting their active error recovery (AER) protocol to Odyssey. Tutorials were given by both groups to exchange information about CANEs slot model programming, Bowman capabilities, and the AER protocol. Currently we are co-developing the AER protocol in the CANEs EE using Bowman as the NodeOS.

Our work continues on the Active push simulation, an effort to extend previous work in wide-area server selection by collecting and sharing path performance data at active nodes.

We continued to work on an admission control algorithm to account for both the processing and link bandwidth requirements for active flows with quality of service requirements. The algorithm models an active flow using Markov modulated fluid sources to represent both traffic arrivals and processing.

Publications and Presentations

Travel

Administrative Issues

Plans for Next Quarter


Ellen Zegura