Expressing and Exploiting Concurrency in Networked Applications in Aspen

This document argues that current programming models are not well-suited for high-performance networked applications that exhibit substantial parallelism among independent tasks, flows, and data elements. In particular, current software architectures impose a daunting tradeoff between programmability and performance, current data communication models violate encapsulation and modularity, and current programming languages provide insufficient primitive support for concurrency and runtime resource management. Aspen is a new language and runtime system built from the ground up to target networked applications with inherent explicit parallelism. Applications in Aspen are expressed as flowcharts consisting of sequential code modules and explicit communication channels; this abstraction allows the programmer to factor out the expression of concurrency and data communication from the core application logic. Additionally, the Aspen runtime dynamically allocates resources to the individual modules and manages all data movement between modules and on the external network, thus freeing the programmer from these concerns. A web server and videoon-demand server coded using Aspen achieve performance competitive with (and sometimes better than) hand-tuned C++ versions despite using 50—80 % fewer lines of program code. 1