Design of an optimal AQM controller for the Diffserv

architecture

Design of an optimal AQM controller for the Diffserv

architecture

Leonardo Rangel / Nelson FonsecaInstitute of Computing - IC

Unicamp, Campinas, 13086-970, SP, Brazil{leonardo.rangel,nfonseca}@ic.unicamp.br

Congestion Control SystemCongestion Control System

Diffserv: Differentiated Services over the Internet

Integrated Congestion Control System● TCP congestion control mechanism can dynamically

adjust the amount of data that a client sends to network;● AQM mechanisms monitor the queue state at the routers,

therefore can detect incipient congestion;● Integrated TCP/AQM Congestion Control System:

AQM role: early notification of incipient congestion to TCP senders;When notified about the incipient congestion, TCP senders can

decrease their transmission rates before queue overflows.

TCP/AQM viewed as a feedback control system:● The congestion level (router queue size - q) is determined

by TCP transmission rate, which is regulated by drop/mark levels;

● An appropriate mark/drop probability (p) is calculated by AQM controller, C(s);

● The proposed controller also stabilizes the system plant, P(s), minimize jitter and give users a certain minimum guaranteed transmission rate.

The feedback control systemThe feedback control system

P(s)

C(s)

q

q0

+

δp

p

-δq

p0

A basic feedback control system

Current statusCurrent status

What we already done:● Modeling of TCP flows and AQM routers behavior;● Investigation of the system's best equilibrium point;● Controller synthesis using Optimization and Control

Theory;● Implementation of the TCP/AQM system in a

continuous-time environment (Simulink/Matlab);● Controller discretization;● Implementation of the controller in a discrete-time

dynamic network environment (NS-2).

Next steps:● Define representative NS-2 simulation scenarios;● Simulation and comparison with current proposals.