Traffic Management for the Available Bit Rate (ABR) Service in ...

good transient performance. Since real networks are in a transient state most of the
time (sources are rarely in nite and ABR capacity varies), the transient performance
ofascheme deployed under these conditions is of importance.
The di erence in approach between the OSU scheme and ERICA is that while
the former attempts to achieve e ciency and fairness one after another, the latter
attempts to move towards e ciency and fairness at the same time. It uses an aggres-
sive core algorithm - the maximum of an \e ciency term" (based on the overload
factor and the source's current rate) and a \fairness term" (based on the available
capacity and the number of active sources). The ERICA schemes still rely on mea-
surement of load, capacity and the number of active sources to calculate the rates.
The ERICA+ scheme attempts to achieve an operating point of 100% utilization and
a target queueing delay. The use of the queueing delay metric allows the scheme to
be robust to errors in measurement and feedback delays (which manifest as queues at
the switch). Simulation results with di erent con gurations and tra c patterns have
also been presented.
Chapter 7 examines the design of source rules in the ATM Tra c Management
framework, i.e., how \open-loop"control complements the \closed-loop" feedback sys-
tem. This dissertation work has helped develop a number of the rules in the inter-
national standard. However, two of these issues are investigated in depth in this
chapter.
The rst issue is the design of \Use-it-or-Lose-it" policies. These policies takeaway
a source's assigned rate if the source does not use it. The choice of the policy has
a signi cant impact on ABR service capabilities, a ecting the performance of bursty
(on-o ) sources and sources bottlenecked below their network-assigned rate. We
376