On Variables as Access Sequences in Parallel Asynchronous Computations

Abstract

We introduce a new method for coordinating access to shared variables in parallel asynchronous computations. The method is based on access sequences, the representation of each shared variable as the sequence of values written to and read from the variable, and on parallel operations, a mechanism for accessing groups of shared variables atomically. Parallel operations on access sequences replace locks as the basis for implementing atomic actions and process synchronization. Advantages include reduction of the potential for deadlock and starvation and greater concurrency in accessing shared memory. The principal costs are the increase in space needed to represent shared variables and in the complexity of the interconnection network and the memory modules. Note: Abstract extracted from PDF file via OCR