ABR Multicast

[With Nazih Elderini, Professor, Dept of Computer Science, Alexandria University, Egypt]

Download Thesis

The point-to-multipoint ABR service within ATM networks is important for many emerging data applications. Examples of multipoint applications include distance learning, server and replicated database synchronization, advertising, searching and data distribution applications. ABR traffic management for point-to-multipoint connections controls the source rate to the minimum rate supported by all the branches of the multicast tree. Feedback consolidation at the branch points becomes a necessary operation to avoid the feedback implosion problem, where the number of backward resource management (BRM) cells received by the source is proportional to the number of leaves in the multicast tree. In addition, the allowed rate of the source should not fluctuate due to the varying feedback received from different leaves. A serious problem may also arise if the BRM sent by the branch point towards the root doesn't consolidate feedback information from all branches. This will introduce noise called "consolidation noise" leading to rate oscillation at the root.

A number of algorithms have been proposed for extending ABR congestion avoidance algorithms to perform feedback consolidation at the branch points. Schemes that attempt to maximize accuracy of feedback information tend to be slow in providing feedback to the source when the conditions in the network change. Accuracy can be traded for speed by having a switch generate feedback information before it has all the necessary information from downstream paths. This can cause consolidation noise, leading to a heavy rate at the source.

In this thesis, we proposed an improved algorithm for feedback consolidation. The new algorithm combines benefits from the previous developed algorithms with reduced overhead. The performance of the proposed algorithm and the previous algorithms is compared under a variety of conditions. Results of the simulation experiments indicate that the algorithm we propose doesn’t suffer from the consolidation noise, while exhibiting a fast transient response with accurate feedback information.

We have also proposed new criteria to control the RM ratio (BRM cells received/FRM cells sent) at the source. The new criteria is very straightforward to implement and can be used by any consolidation algorithm that permit sending extra fast overload feedback cells. Results have shown that the new method has the effect of accelerating the convergence of the RM ratio to one while not affecting the transient response.

Thesis