Abstract state machines for data-parallel computing

Abstract

The current programming paradigm for data-parallel computations is shifting with the rapidly increasing data growth on the web. It gives programmers more challenges than ever before. In this paper we propose Parallel Abstract State Machines (P-ASMs) that can empower programmers, no matter how experienced, by providing a well-founded systems engineering method to model data-parallel computations at arbitrary levels of abstraction. Particularly, we focus on discussing how P-ASMs can capture two classes of data-parallel computations that are most important in practice - ones that are always-consistent and ones that require transactional data consistency.