الفهرس 
chapter 1Only 14 pages are availabe for public view
 |  | from | 269 |  |  |
| | | from | 269 | | |
Abstract
ATM network architecture is capable of supporting a wide range of
connections with different quality of service requirements and traffic
characteristics. This dynamic environment creates difficult traffic control
problems when trying to achieve efficient use of network resources. One of
such problem 1s buffer allocation. in this thesis, the buffer allocation problem
in A TM is handled in two stages: classification of the incoming traffic into
multiple priority classes, and then managing the buffer in order to guarantee
QoS of an these classes.
Since ATM traffic characteristics are quite diverse, and quality of
service (QoS) and bandwidth requirements vary considerably, the Analytic
Hierarchy Process (AHP) is proposed as a new solution to classify the A TM
traffic into three priority classes according to many parameters. These
parameters are QoS requirements (i.e .• cell loss and delay), burstiness, coding
and compression methods, holding time, and bandwidth. After the
classification process, the input buffer of the A TM switch is partitioned by two
fixed thresholds in order to assign these three service classes and then the
switch is analyzed_ the numerical results confirm an improvement over the
traditional one threshold scheme.
While this conventional solution achieves good switch performance. it
can not dynamically regulate traffic flows according to changing network
conditions, So. a fuzzy logic solution is suggested since it can deal with real
world imprecision. A fuzzy buffer management controller (FBMC} forr A TM
networks th.llt handles both the two sides of the problem is proposed in two
parts: fuzzy classifier and fuzzy threshold controller. fn the fuzzy classifier. the
traffic is categorized based on delay and cell Joss into three priority classes that
incorporate seven suhc1asses in order to realize accurate classification w1th
respect to both loss and delay. In the fuzzy threshold controller, the two
dynamic thresholds of the input buffer are indicated based on the cell arrival
rates of these three classes in order to prevent congestion as soon as possible.
Simulation results indicated that the proposed FDMC achieves higher
throughput by about 4% over the traditional controller and achieves satisfactory
QoS of all traffic classes in all cases of traffic models.
However, no clear and general technique has been presented to map
existing knowledge on traffic control onto the design parameters of the fuzzy
logic controller. Self-leaning capability of the neural network could be
deployed in the fuzzy logic controller in order to simplify the design procedure
and obtain better control results. So, neurofuzzy buffer management controller
(NFSMC) is proposed that also incorporate classifier and threshold Controller
The comparison between the FBMC and the NFBMC shows that in most
experiments, the NFBMC results in fewer cell discards and higher throughput.
Although the Improvement of the results is lower than 1%, using a neurofuzzy
approach is preferred since it automatically selects the best membership
functions and rules instead of using a trial anti error procedure In tl1e fuzzy
controller.