Date: Wednesday, 15-Jan-97 00:19:51 GMT Server: NCSA/1.1 MIME-version: 1.0 Content-type: text/html Last-modified: Monday, 27-Mar-95 17:11:58 GMT A CATV Based Metropolitan Area Network Using Distributed Switching

Arthur I. Karshmer and Renwei Yan

Computer Science Department
Computing Research Laboratory
New Mexico State University
Las Cruces, NM 88003 USA

ABSTRACT

A number of protocols have been proposed to allow the use of standard cable television plants to support metropolitan area networking. In the present work we describe a new approach to solving the problem that has some interesting properties in its ability to support large numbers of users at fixed transmission rates at a relatively low cost.

Introduction

A major goal of all computer network designs is the distribution of information in a fast, error free manor at the lowest possible cost. One method of accomplishing this goal is to use already existing technologies in new and imaginative ways. Although designed for different purposes, the cable television systems (CATV) already in place in much of the U.S. provide an interesting and potentially useful platform for the distribution of digital data on a metropolitan area scale. As the CATV systems in use were initially designed for one way communications, any plan to use them must include either upgrading currently installed upstream amplifiers or the actual implementation of them.

The problems associated with using these types of facilities were studied and it has been demonstrated that using such facilities is technically and economically feasible [Karshmer, Thomas and Phelan - 1991; Karshmer and Thomas - 1989]. These studies along with others [Hatamain - 1985; Hafez - 1985] have demonstrated potential strengths and weaknesses in using such systems. Possibly the greatest strength of using CATV based networks is their almost universal existence in the U.S. Approximately 60% of all homes are currently connected to a cable plant, and virtually every building in most cities is "passed" by a cable.

One of the key problems associated with using the branching tree topology of a cable plant is access control. Due to the separation of upstream and downstream channels, any message from a source node (a leaf) must be sent upstream to the head end and then downstream to the destination node (another leaf). This round-trip time can cause a large delay message transmission and reception which can render an access control technique such as CSMA/CD useless.

In the current work, we present results from the analysis of a system that reduces message transmission time on a two-way CATV network while not decreasing network throughput. We develop a model of a two-level CATV networking scheme which is then tested using well know simulation techniques. Finally, the simulation results are analyzed and are compared to best throughput and expected delay times.

Distributed switching over a single-trunk CATV network

Figure 1 shows a simplified view of our two-level, single-trunk CATV network with M switches installed on the trunk. Each switch is in turn connected to a branch cable that supports K branch nodes (network subscribers). Each trunk node (switch) has three inputs (trunk upstream, trunk downstream and branch upstream) as well as three outputs. Each branch node has one output for branch upstream and one input for branch downstreamSince all trunk nodes have a certain level of intelligence, the traffic flow on the trunk is totally controlled. The goal of the intelligent trunk nodes is to turn-around messages onto the appropriate branch cable at the earliest possible time. The head end is therefore not the only node to that can redirect messages. If our system is able to redirect messages at the earliest possible time, we believe the system can reduce message travel time, reduce message delay and increase system throughput.

Figure 2 shows the simulation results for the 512 (M * K) node case with different node distribution on the trunk and branch cables. The distributions are 1 trunk node and 512 branch nodes, 5 trunk nodes and103 nodes per branch, 10 trunk nodes and 51 nodes per branch, 20 trunk nodes and 25 nodes per branch and 40 trunk nodes and 13 nodes per branch. Figure 5 shows the DELAYs related to these different distributions. The following figure shows a packet length of 512 or 2048-bits and a baud rate of 1Mbps.

Figure 2 shows that with a fixed baud rate, (1Mbps) and a fixed packet size (512 or 2048 bits), the throughput (total number of packets) on each branch cable is almost identical. Therefore, the total throughput of the system within a given time is expressed in the following equation.

As the value, Num_Packets_per_branch is almost identical to the same values for G and S on each branch cable, it seems clear that putting the control node(s) on the trunk cable will linearly increase the throughput for the entire network. In this case, the implementation of an intelligent node on the trunk cable resulted in the desired performance. Since Slotted Aloha can double the throughput on each branch, we can see that the total throughput is also doubled. This implies that the better the protocol on the branch level, the more the total system throughput increases.

Figure 3 shows that the DELAY time may be reduced even if the number of trunk control nodes increases. The following equation shows that the de lay is mainly caused by two factors: the size of waiting queue and the retransmission rate (G), which is what queuing theory would predict. If there are too many messages at the trunk control nodes waiting to be transmitted, the DELAY will increase. This means that if the traffic is greater than trunk capacity, the store and forward protocol on truck level will cause a long DELAY and deteriorate overall system performance .

Conclusions

The results from the current study lead us to a number of conclusions regarding the use of a two-level distributed intelligence approach to using CATV equipment to support metropolitan area networking. Briefly stated, they are:

Adding control nodes on the trunk cable (increasing M) will reduce the DELAY with the number of uses (subscribers) in the system being fixed.

Proper values for the average transmission success rate (S) and average transmission attempts (G) will also tend to increase the throughput.

Increasing the Baud rate, tends to reduce the Packet_time, and therefore has a positive effect in reducing the Waiting_queue_SIZE, which in turn reduces the DELAY (see Figure 5).

Interestingly, increasing the Baud rate beyond a certain limit tends to cause the Waiting_queue_SIZE to increase. Consequently the Packet_time is reduced and the DELAY time increases (see Figure 5).

Essentially, our ability to reduce the DELAY time is a function of the trunk cable's transmit rate (bit rate) and the Packet_Length. Increasing the value of M will not cause the DELAY time to be increased because the throughput on each branch is effectively fixed and has a normal distribution probability. Therefore, the more nodes (users) on a branch, the less frequently each branch node can send a Packet onto branch cable as the throughput on each branch is fixed. As a result, to the extent that a trunk node can handle the traffic, the waiting queue size will not increase and the DELAY time will not increase.

The model clearly demonstrates that our design is extendable: we can increase the number of users by simply adding trunk control nodes to increase the system throughput without increasing the average system DELAY. This is an important result as it tells us that a small number of trunk nodes, which are inexpensive and simple to install, have a dramatic impact on overall network performance. Our results were obtained using a Baud rate of 1Mbps - using a more realistic figure in the range of 4Mbps could only enhance our findings.