Determination Of End-To-End Delays Of Switched Ethernet Local Area Networks

ABSTRACT

The design of switched local area networks in practice has largely been based on heuristics and experience; in fact, in many situations, no network design is carried out, but only network installation (network cabling and nodes/equipment placements). This has resulted in local area networks that are sluggish, and that fail to satisfy the users that are connected to such networks in terms of speed of uploading and downloading of information, when, a user’s computer is in a communication session with other computers or host machines that are attached to the local area network or with switching devices that connect the local area network to wide area networks. Therefore, the need to provide deterministic guarantees on the delays of packets’ flows when designing switched local area networks has led to the need for analytic and formal basis for designing such networks. This is because, if the maximum packet delay between any two nodes of a network is not known, it is impossible to provide a deterministic guarantee of worst case response time of packets’ flows. This is the problem that this research work set out to solve. A model of a packet switch was developed, with which the maximum delay for a packet to cross any N-ports packet switch can be calculated. The maximum packet delay value provided by this model was compared from the point of view of practical reality to values that were obtained from literature, and was found to be by far a more realistic value. An algorithm with which network design engineers can generate optimum network designs in terms of installed network switches and attached number of hosts while respecting specified maximum end-to-end delay constraints was developed. This work revealed that the widely held notion in the literature as regards origin-destination pairs of hosts enumeration for end-to-end delay computation appears to be wrong in the context of switched local area networks. We have for the first time shown how this enumeration should be done. It has also been empirically shown in this work that the number of hosts that can be attached to any switched local area network is actually bounded by the number of ports in the switches of which the network is composed. Computed numerical values of maximum end-to-end delays using the developed model and algorithm further revealed that the predominant cause of delay (sluggishness) in switched local area networks is the queuing delay, and not the number of users (hosts) that are connected to the networks. The fact that a switched local area network becomes slow as more users are logged on to it is as a result of the flow of bursty traffic (uploading and downloading of high-bit rates and bandwidth consuming applications). We have also implemented this work’s model and algorithms in a developed C programming language-based inter-active switched local area networks’ design application program. Further studies were recommended on the need to develop method(s) for determining the maximum amount of traffic that can arrive to a switch in a burst, on the need for the introduction of weighting function(s) in the end-to-end delay computation models; and on the need to introduce cost variables in determining the optimal Internet access device input and output rates specifications.