Browse Reports

In this dissertation a new approach to the synchronization of accesses to shared data objects is developed. Traditional approaches to the synchronization problems of shared data accessed by concurrently running computations have relied on mutual exclusion -- the ability of one computation to stop the execution of other computations that might access or change shared data accessed by that computation. Our approach is quite different. We regard an object that is modifiable as a sequence of immutable versions; each version is the state of the object after an update is made to the object. Synchronization can then be treated as a mechanism for naming versions to be read and for defining where in the sequence of versions the version resulting from some update should be placed. In systems based on mutual exclusion, the timing of accesses selects the versions accessed. In the system developed here, called NAMOS, versions have two component names consisting of the name of an object and a pseudo-time, the name of the system state to which the version belongs. By giving programs control over the pseudo-time in which an access is made, synchronization of access to multiple objects is simplified. NAMOS is intended to be used in an environment where unreliable components, such as communication lines and processors, and autonomous control of resources occasionally cause certain objects to become inaccessible, perhaps in the middle of an atomic transaction. Computations may also suddenly halt (perhaps as the result of a system crash) never to be restarted. NAMOS provides facilities for recovering from such sudden failures, grouping updates into sets called possibilities, such that failure of any update belonging to a possibility prevents all of the other updates in the possibility. The naming mechanism of NAMOS also provides a useful tool for restoring a consistent state of the system after a failure resulting in irrecoverable loss of information or a user mistake resulting in an inconsistent state. An important motivation for the development of NAMOS is the need to support decentralized development of application systems by combining existing application systems that deal with shared data. NAMOS supports the construction of modules that locally ensure their own correct synchronization and recovery from inaccessibility. Larger modules that use several separately designed modules can then be constructed, perhaps with additional synchronization constraints, without modifying the modules used. In most systems based on mutual exclusion, such post hoc integration of modules is difficult or impossible.