Optimistic Programming in PVM

Optimism is a powerful technique for increasing concurrency. A program can gain concurrency by making an optimistic assumption about its future state, and verifying the assumption in parallel with computations based on the optimistic assumption. To date, use of optimism has been restricted to specialized systems due to the diiculty of writing optimistic programs. In this paper, we deene and justify the deenition of optimism as any computation that uses rollback. Optimism is eeective in avoiding the latency of predictable computations. Performing computations remotely incurs the additional latency of communications time, and so remote computations make good candidates for the application of optimism. By assuming the outcome of a remote computation, a task can proceed in parallel with connrmation of the guess. By guessing correctly, a task can save the amount of time that remote connrmation would have required. HOPE is our model of optimistic computation built upon these notions. HOPE consists of one data type and four operations that provide for the concise and portable expression of optimistic algorithms. HOPE can be embedded in any system providing concurrent task that communicate by exchanging messages. This paper describes our current eeort to construct a HOPE prototype on top of PVM. The basic requirements are to integrate PVM with a task checkpoint and rollback mechanism, make the PVM 1 message queues robust with respect to task rollback, and nally build the HOPE prim-itives on top of this rollback mechanism. We present the services provided by the resultant system, and describe the current state of the prototype.