Functional programming abstractions for weakly consistent systems

Sivaramakrishnan Krishnamoorthy Chandrasekaran, Purdue University

Abstract

In recent years, there has been a wide-spread adoption of both multicore and cloud computing. Traditionally, concurrent programmers have relied on the underlying system providing strong memory consistency, where there is a semblance of concurrent tasks operating over a shared global address space. However, providing scalable strong consistency guarantees as the scale of the system grows is an increasingly difficult endeavor. In a multicore setting, the increasing complexity and the lack of scalability of hardware mechanisms such as cache coherence deters scalable strong consistency. In geo-distributed compute clouds, the availability concerns in the presence of partial failures prohibit strong consistency. Hence, modern multicore and cloud computing platforms eschew strong consistency in favor of weakly consistent memory, where each task's memory view is incomparable with the other tasks. As a result, programmers on these platforms must tackle the full complexity of concurrent programming for an asynchronous distributed system. This dissertation argues that functional programming language abstractions can simplify scalable concurrent programming for weakly consistent systems. Functional programming espouses mutation-free programming, and rare mutations when present are explicit in their types. By controlling and explicitly reasoning about shared state mutations, functional abstractions simplify concurrent programming. Building upon this intuition, this dissertation presents three major contributions, each focused on addressing a particular challenge associated with weakly consistent loosely coupled systems. First, it describes A NERIS, a concurrent functional programming language and runtime for the Intel Single-chip Cloud Computer, and shows how to provide an efficient cache coherent virtual address space on top of a non cache coherent multicore architecture. Next, it describes RxCML, a distributed extension of MULTIMLTON and shows that, with the help of speculative execution, synchronous communication can be utilized as an efficient abstraction for programming asynchronous distributed systems. Finally, it presents QUELEA, a programming system for eventually consistent distributed stores, and shows that the choice of correct consistency level for replicated data type operations and transactions can be automated with the help of high-level declarative contracts.

Degree

Ph.D.

Advisors

Jagannathan, Purdue University.

Subject Area

Computer Engineering|Computer science

COinS