ACD Atlas Computing Division Distributed Computing Systems

Jump To Main Content

Jump Over Banner

Home

ACDDCSProjects

Jump Over Left Menu

DCS Projects: University of Strathclyde

D HUTCHISON and W D SHEPHERD

A FEASIBILITY STUDY INTO THE COMPLEXITY OF GATEWAYS FOR CONNECTING ETHERNET TO RING NETWORKS

Oct 80 - Sept 81

Over the past few years there has been an increasing interest in the design and construction of local area networks based on high bandwidth communication through thin wires. The two dominant architectures used are the broadcasting and loop or ring system. Two well known examples of these are the Ethernet system developed at Xerox Palo Alto and the Cambridge Ring respectively.

Originally it was thought that local area networks would consist of hundreds of devices connected to a single communication medium. It now seems more likely that any establishment will have a number of local area networks with tens of devices attached to them rather than hundreds, serving the particular needs of various departments. Already there are appearing on the market personal computers complete with operating system and hardware interface which allow them to be linked together using a particular communication architecture. In the near future more of these devices are going to become available, especially in the areas of office systems, laboratories and word processing. It appears extremely likely that the communication systems will be either Ethernet or Ring based, but it is extremely unlikely that one approach will dominate the other. This means that not only will there be a number of local area networks in an establishment but some will be Ethernets and others Rings.

In order to obtain the best use of resources it will be necessary in some cases to link networks together by means of a gateway or bridge. In the general case a gateway will have to be able to deal with the linking of an Ethernet system to a Ring system.

Strathclyde are carrying out a feasibility study into the complexity involved in the design and construction of gateways between local area networks based on Ethernet and Ring communications, respectively. In particular they are considering the linking of a Cambridge Ring System to the Strathclyde Network System (the latter being built in the Department of Computer Science at the University of Strathclyde with the aid of a grant from the Carnegie Trust).

The project is expected to produce design guidelines for gateway implementations; in particular the investigators expect to be able to construct hardware and software for an Ethernet-Ring gateway.

An award has been made to Mr Shephard for the construction phase of this project entitled Gateways for the Interconnection of Cambridge Rings and Ethernet-like Networks.

DR D G SMITH

OVERLOAD CONTROL IN PROCESSOR-BASED SYSTEMS

Feb 83 - Jan 84

Aims

This project aims to investigate methods of controlling the behaviour of processor-based systems under overload. An algorithm will be studied which observes the system and controls the degradation of service as the loading increases.

Background

In any information processing system there is always the problem of assigning resources of the system to users who are demanding service in a satisfactory fashion. The questions of what resource assignment policies one can adopt, and of what is the measure of satisfaction by which they should be compared depends, on the design of the information processing system and the purpose that it was designed to serve. But it seems likely that similar techniques might be used to study quite a variety of structures, and that it would be possible to construct a kind of theory of overload control by considering the common aspects of a number of different problems.

There are two broad areas into which the field can be divided, these being data transmission networks and data processing systems. In the transmission networks principal concern is with the transmission and reception of information, and, while it may be necessary to examine some of the information and do processing work on it at the nodes, this is incidental to the main function of the network which is the transmission. In the data processing system the main purpose is information processing, which is undertaken to achieve some end in computing or real time control. There may be transmission of information over various physical facilities in the course of performing the processing function, but this time the transmission is incidental to the main function of the system.

Both kinds of structure need some sort of load control. At the present time. more allention has been given to flow control in a data transmission network than to load control in an information processing system. But neither application has yet produced a well founded framework of definitions and objectives in terms of which the load control problem can be discussed.

Present State of the Art

The information processing function which the investigator has in mind is the control of an exchange in an SPC telephone system. There are a number of such systems currently operating throughout the world. Each different system has a different structure for the processing, and probably they all have different load control arrangements heuristically designed in a way which will, (it is hoped) to some extent meet a variety of objectives which are judged to be desirable.

The situation with data-networks is rather different. The objectives are clearer, namely to avoid any kind of lockout or crash, while minimising the response time of the network to accepted traffic. Load control methods have been in use and under discussion for some time, with a view to achieving these ends.