| dc.description.abstract | This research work deals with a new design/ and study of a token
ring local area network (LAN) for small systems application
like microcomputer network. The salient features of the work
carried out are as follows:
“ Development of a new access technique and a distributed
algorithm to realize a token ring/
“ hardware design and construction of a microprocessor
based Ring Interface Unit and an experimental ring net,
- construction of a traffic-adaptive simulation model with a
cyclic queuing procedure for performance evaluation/ and
- study of a multiple token ring which is modelled as
separate single rings with a mixed class of stations.
Many access techniques have been developed to satisfy the
requirements of a local area network/ but the token ring has
turned out to be the most suitable scheme for different applications
ranging from office automation to real-time systems like
process control. Numerous comparative analyses have proved the
performance superiority of the token ring over the other schemes.
The analysis has also shown the robustness of the token ring to
the large variation in the network parameter and traffic.
Elowever/ as far as small systems applications are concerned/ the
realization of a token ring becomes quite complex in its control
procedure/ and relatively expensive mainly due to lack of an LSI
implementation of this access scheme. In this work/ a new and simple approach is described for designing and implementing an
inexpensive token ring system. Also/ modelling and performance
studies of the ring net are done in a more realistic small
systems environment.
Initially/ in order to find a cost-effective approach to
construct a ring node, standard Data Link protocols that are
readily available in the form of low-cost single chip hardware
are surveyed. As a result/ the IBM-SDLC loop protocol is found to
be suitable due to the resemblance of its architecture with the
token ring. However/ in the SDLC loop architecture/ the
procedure of having a permanent primary station to control the
entire loop generates an unbalanced traffic in the network; and
poses reliability problems if the primary station fails. These
are unsuitable features for a multi-user system like LAN.
Therefore/ to avoid these difficulties, in this work/ all the
SDLC stations are given an equal priority to control the loop-
This is done by suitably modifying the SDLC frame format and the
polling procedure.
In accordance with these modifications/ a distributed
algorithm is developed to convert the centralized SDLC loop
architecture into a decentralized network like that of token
ring. This algorithm allows the role of the primary station to
circulate among all the stations in the loop. That is, it permits
any station to acquire this primary status when it has a message
to transmit. After the loop-transmit/ the transmitting station
is made to pass its primary status to the next access seeking
station/ and thus making the loop control become distributed. To
realize a loop station/ an INTEL 8085 microprocessor based Ring
Interface Unit is constructed and it is tested in an experimental
ring net. The network-control part of the system includes a protocol controller/ a DMA controller/ a transceive buffer, a
clock recovery and a bypass circuits/ and network command and
control software.
Although there exist numerous performance analyses on
token ring, analyses pertaining to the study of the system in a
realistic environment are scarce. In this work, a more general
and versatile computer simulation model with a cyclic queueing
procedure is constructed. In this model, it is observed that
working with a single model (explicit OR implicit polling type)
to operate a wide range (0.1 to >0.9) of network traffic is
inefficient. For instance, in the case of a low message arrival
rate (<0.5), the execution time of the simulation program (of
explicit polling) becomes prohibitively high. Because, a light
traffic in the ring induces the circulation of idle-token for
most of the time: a system state that does not contribute to the
servicing of a message. Therefore, in order to improve the
program's runtime efficiency, the model is broken into two
.different types - low traffic and high traffic models. Depending
on the computational time of the previous run, the proper model
is switched on for further simulation in the other ranges of
traffic. This "traffic-adaptive" simulation model reduces the
overall CPU-time of the program for all ranges of traffic.
Attempts are made to verify this simulation model by comparing it
with known analytical results, and the results are found to be
satisfactory. Apart from this varying traffic model, a saturated
load (continuously queued) model is also developed. This aims to
analyse the sensitivity of the token ring at extreme operational
conditions.
The simulaton work is largely directed towards the modelling
of the ring net in a more realistic small systems environment,
i.e., where the network is connected with a mixed class of stations like work-station/ file-server/ and bridge. This
unbalanced traffic model is used to analyse multiple rings (2-
rings) which are connected by a bridge. The flow of message
between rings is characterized by an *'interaction-probability"
with which a ring net transfers its messages to the other. This
parameter is used to calculate the arrival rate of messages at a
bridge-station of the ring net, and the multiple ring net is then
modelled as logically separated single rings which have an
unbalanced traffic in the net due to the presence of a bridge
station. While doing so, the delay history of the messages that
arrive at the bridge-station from the other ring is properly
accounted for.
The above analyses are mainly used to address the problems
and to device effective solutions to cope with the natural growth
of stations in a ring net of a small systems environment. From
different analyses^ it appears that configuring a large single
ring net into multiple rings connected by a bridge is a better
choice than altering the existing capacity of the growing ring.
Besidesf in order to improve the throughput-delay efficiency of
the multiple ring net, performance studies for various ringinteraction-
probablities are done. The different techniques
applied in this effort are:
a) giving a "priority-servicing" at the bridge-station/ and
b) applying a "resource-balancing" approach in the ring nets
to reduce the flow of many messages accross the bridge. | |
