Research With OPNET: Multi-layer Restoration

The impact of outages in conventional telephony networks is very well understood. Data
link technologies like SONET have very specific targets for the time interval within
which restoration must be achieved. Missing the restoration time target means that an
outage becomes visible at higher layers, and will likely spread rapidly over a service
area. This knowledge is missing for data communications networks, including those like
the Internet, which are based on TCP/IP.

The goal of our project is to measure the effects of outages of varying durations to determine whether there is a smoothly graded response, or whether there are specific time horizons, which are critical. With OPNET, we will measure end-to-end effects at several layers in the protocol stack, from the data link layer up to the application layer.

These results will suggest the types of restoration mechanisms that are useful in IP
networks. They will also highlight those places in the protocol stack where restoration
functionality might be incorporated to best effect. For example, local keep-alives for
a specified hold off period might be useful in keeping all the TCP connections at a
node alive while restoration is effected at the physical or data link layer. Once these
results are available, we will be able to go on to design and implement such mechanisms.

Our vision is that the final proposed mechanism might be "Multi-layer Restoration", i.e., 
multiple layers in the protocol stack should coordinate with each other in order to get the best restoration performance. Of course, the proper mechanisms depend on our first-step measurements. 

Publication:

Qiang Ye and Mike MacGregor, "On the Resilience of SACK  and Reno TCP",  Fourth International Workshop on the Design of Reliable Communication Networks (DRCN 2003), Banff, Alberta, Canada. October, 2003.

Qiang Ye and Mike MacGregor, "SACK TCP Resilience Improvement with OPNET", OPNETWORK 2004, Washington D.C. Aug. 30-Sept. 3, 2004 (Submitted)