43 - Distributed Algol

Summary

A distributed computation is one that takes place at many different sites, each of which controls some resources located at that site. For example, the sites might be nodes on a network, and a resource might be a device or sensor located at that site or a database controlled by that site. Only programs that execute at a particular site may access the resources situated at that site. Consequently, command execution always takes place at a particular site, called the locus of execution. Access to resources at a remote site from a local site is achieved by moving the locus of execution to the remote site, running code to access the local resource, and returning a value to the local site.

In this chapter we consider the language ℒ{nat cmd ⇀∥}, an extension of Concurrent Algol with a spatial type system that mediates access to located resources on a network. The type safety theorem ensures that all accesses to a resource controlled by a site are through a program executing at that site, even though references to local resources may be freely passed around to other sites on the network. The key idea is that channels and events are located at a particular site and that synchronization on an event may occur only at the site appropriate to that event. Issues of concurrency, which are to do with nondeterministic composition, are thereby cleanly separated from those of distribution, which are to do with the locality of resources on a network.