Efficient Software Implementation of Stream Programs

Abstract: The way we use computers and mobile phones today requires large amounts of processing of data streams. Examples include digital signal processing for wireless transmission, audio and video coding for recording and watching videos, and noise reduction for the phone calls. These tasks can be performed by stream programs—computer programs that process streams of data. Stream programs can be composed of other stream programs. Components of a composition are connected in a network, i.e. the output streams of one component are sent as input streams to other components. The components, that perform the actual computation, are called kernels. They can be described in different styles and programming languages. There are also formal models for describing the kernels and the networks. One such model is the actor machine.This dissertation evaluates the actor machine, how it facilitates creating efficient software implementation of stream programs. The evaluation is divided into four aspects: (1) analyzability of its structure, (2) generality in what languages and styles it can express, (3) efficient implementation of kernels, and (4) efficient implementation of networks. This dissertation demonstrates all four aspects through implementation and evaluation of a stream program compiler based on actor machines.