[Colloq] **Ph.D. Thesis Defense ** Tuesday, May 20, 9:30AM**

[Rachel Bates]

 Ph.D Thesis Defense

Date: Tuesday, May 20
Time: 9:30 - 11:00 AM
Place: 149 Cullinane Hall

Title: On the Efficiency and Fairness of AIMD-based Congestion Avoidance
       Algorithms

by

Adrian Lahanas


Advisor:   V. Tsaoussidis
Committee: Gene Cooperman, Paul Attie
External Committee Member: Ibrahim Matta (BU)

Abstract

Congestion is a common problem in packet networks. Several resource
management techniques have been designed to deal with congestion. AIMD
is the dominant algorithm for congestion control since it is the core
mechanism of TCP.

I will discuss the key components of AIMD that achieve efficiency and
fairness. I will exploit a property of all flows within a network to
share common knowledge in a distributed manner. Based on this property,
I will discuss the design of Additive Increase Multiplicative Decrease -
Fast Convergence (AIMD-FC).

AIMD-FC achieves better efficiency and fairness than AIMD. An
AIMD-FC-based system converges to fairness faster than an AIMD-based
system. In addition, bandwidth utilization of the participating flows
is significantly higher. I will demonstrate the ease of its deployment
and I will discuss its practical impact.

Finally, I will present some AIMD-FC-derived algorithms that improve
efficiency and fairness further.  For example, AIMD-FC+ improves
efficiency of AIMD by 14%. t-AIMD and Rate-AIMD improve fairness of
AIMD while achieving the same efficiency.  The last algorithm, O-1-AIMD,
converges to fairness much faster than AIMD and the other algorithms. The
properties of these algorithms will be proved theoretically and claims
will be supported by experimental results.