Configuring distributed systems

ABSTRACT

Distributed systems are constructed from a set of computational components which are bound together to interact and communicate to achieve some over all objective. In this paper, we examine the issues related to system configuration, particularly those related to component binding. We argue that there is a need for tools to aid the initial construction and subsequent management of the structure of distributed systems.Currently, distributed systems are most commonly structured using a client/server approach. Servers register the service they provide by installing the name and type of the service together with a reference to that service in a distinguished server generally known as a name server. Clients bind to a particular service by looking up the reference in a name server (the reference to which must be well known). Once the client has retrieved the service reference, the service may be directly invoked (although, if the reference is location independent, it may require translation to an address). ANSA[1] is an example of a distributed systems environment which conforms to this paradigm with the extension of providing a more general service than name to reference translation via its Traders. Traders provide the ability to locate services by attribute in addition to simple name lookup.Problems with the client/server approach arise when the systems we wish to construct have a more complex interconnection structure than the simple many-to-one interconnection pattern which the approach directly supports. In particular, where the distributed program is more sensibly viewed as a set of agents with peer-to-peer connections. An example is the way applications may be constructed under Unix from a pipeline of filter processes. Although all reasonable systems allow servers to be the client for other services, the interconnection process becomes tortuous. It is the authors view that as distributed systems mature and become more complex, as servers themselves are realised by distributed multi-agent programs, this problem will become more acute.

REFERENCES

• 1: ANSAware Release 3.0 Reference Manual, Architecture Projects Management Ltd, Poseidon House, Castle Park, Cambridge, CB3 0RD, 1991.
• 2: Jeff Magee , Morris Sloman, Constructing Distributed Systems in Conic, IEEE Transactions on Software Engineering, v.15 n.6, p.663-675, June 1989 [doi>10.1109/32.24720]
• 3: J. Magee, N. Dulay and J. Kramer, "Structuring Parallel and Distributed Programs", Proceedings of IEE International Workshop on Configurable Distributed Systems, Imperial College, March 1992.
• 4: REX Technical Annexe (1989), ESPRIT Project 2080, European Economic Commission.
• 5: F. Christian, " Automatic Reconfiguration in the Presence of Failures", Proceedings of IEE International Workshop on Configurable Distributed Systems, Imperial College, March 1992.