Naming issues in the design of transparently distributed operating systems

Abstract

Naming is of fundamental importance in the design of transparently distributed operating systems. A transparently distributed operating system should be functionally equivalent to the systems of which it is composed. In particular, the names of remote objects should be indistinguishable from the names oflocal objects. In this thesis we explore the implication that this recursive notion of transparency has for the naming mechanisms provided by an operating system. In particular, we show that a recursive naming system is more readily extensible than a flat naming system by demonstrating that it is in precisely those areas in which a system is not recursive that transparency is hardest to achieve. However, this is not so much a problem of distribution so much as a problem of scale. A system which does not scale well internally will not extend well to a distributed system. Building a distributed system out of existing systems involves joining the name spaces of the individual systems together. When combining name spaces it is important to preserve the identity of individual objects. Although unique identifiers may be used to distinguish objects within a single name space, we argue that it is difficult if not impossible in practice to guarantee the uniqueness of such identifiers between name spaces. Instead, we explore the possibility of Using hierarchical identifiers, unique only within a localised context. However, We show that such identifiers cannot be used in an arbitrary naming graph without compromising the notion of identity and hence violating the semantics of the underlying system. The only alternative is to sacrifice a deterministic notion of identity by using random identifiers to approximate global uniqueness with a know probability of failure (which can be made arbitrarily small if the overall size of the system is known in advance).