Technical Report 93 - 14 July 1993 Randy

Speculative computing is a technique to improve the execution time of certain applications by starting some computations before it is known that the computations are required. A speculative computation will eventually become mandatory (i.e. required) or irrelevant (i.e. not required). In the absence of side e ects irrelevant computations may be aborted. However, with side e ects a computation which is irrelevant for the value it produces may still be relevant for the side e ects it performs. One problem that can result is the relevant synchronization problem wherein one computation requires some side e ect event (a \relevant synchronization") to be performed by another computation, which might be aborted, before the rst computation can make progress. Another problem that can arise is the preemptive delay problem wherein a computation that will perform some awaited side e ect event is preempted by a computation whose importance (e.g. priority) is less than that of computations waiting for the event. In this paper we show how the sponsor model developed for speculative computation in Multilisp can be extended to provide a novel solution to these two problems. The idea is for the computation awaiting some action, such as the production of a value or the release of a semaphore, to sponsor the computation or set of computations that will perform the awaited action. This sponsorship ensures that the awaited action executes, and executes with at least the waiter's level of importance. We show how to apply this technique to solve the above problems for several producer/consumer and semaphore applications. The idea extends naturally to other synchronization mechanisms. To be published in Proceedings of the 1992 Parallel Symbolic Computing Workshop at M.I.T. in Springer-Verlag Lecture Notes on Computer Science, November 1993. This work may not be copied or reproduced in whole or in part for any commercial purpose. Permission to copy in whole or in part without payment of fee is granted for nonpro t educational and research purposes provided that all such whole or partial copies include the following: a notice that such copying is by permission of Mitsubishi Electric Research Laboratories of Cambridge, Massachusetts; an acknowledgment of the authors and individual contributions to the work; and all applicable portions of the copyright notice. Copying, reproduction, or republishing for any other purpose shall require a license with payment of fee to Mitsubishi Electric Research Laboratories. All rights reserved. Copyright c Mitsubishi Electric Research Laboratories, 1993 201 Broadway, Cambridge, Massachusetts 02139 Publication History:{ 1. First printing, MN92-29, October 6, 1992. Presented at Parallel Symbolic Computing Workshop at M.I.T. 2. Revision, MN92-29a, October 21, 1992 3. Revision, TR93-14, July 1993 Extending the Multilisp Sponsor Model July 1993 1