Prof Manohar N. Murthi
Department of Electrical and Computer Engineering
University of Miami

Network bit-rate resource allocation in high-speed networks: maximizing
network efficiency, and an approach to supporting signal processing applications

Abstract

High-speed packet-switched networks such as the Internet are increasingly
being utilized to support increasingly complex real-time applications, 
including signal processing applications.  Such trends necessitate research 
on: (1) design of bit-rate/congestion control to maximize network 
efficiency and thereby maximize the use of bandwidth on increasingly faster 
network links; (2) design of methods for allowing the co-existence of 
sophisticated signal processing and data fusion algorithms on shared 
networks.  In the first part of the talk, we address the design of core 
network bit-rate/congestion control and describe Normalized Queueing Delay 
(NQD), a new concept in congestion control.  By utilizing NQD, which 
combines both delay and packet marking feedback from routers to measure the 
network congestion state, a data-emitting source is able to scale its rate 
dynamically to prevailing network conditions, leading to fair and stable 
rates with nearly full link utilization, and minimal queue oscillation.  In 
fact, the NQD-based D+M TCP (Delay + Marking TCP) achieves better 
performance than FAST TCP, HSTCP, BIC TCP, TCP Vegas, STCP, and the current 
Internet standard TCP Reno.  In the second part of the talk, we examine an 
approach to implementing canonical signal processing and data fusion 
algorithms in shared networks, such as the Internet.  In particular, we 
address a very basic question: at what bit-rates should sensors send their 
measurements to a tracking fusion center in a shared network?  Clearly, 
sensors cannot transmit at arbitrary rates, which could lead to network 
collapse, and current Internet protocols may not lead to desirable tracking 
performance.  To address this problem, we outline a method in which the 
Kalman-filter based multi-sensor target-tracking Quality of Service (QoS) 
function is derived and expressed as a function of the sensor bit-rates.  
With this tracking QoS function, one can derive target-tracking rate 
resource allocation controllers that maximize target tracking QoS while 
allowing for other applications to prosper in a shared network. In 
simulation studies, the new rate control algorithm engenders significantly 
better tracking performance than a standard rate control method.