PhD Thesis: Brandon Hamilton

Citation:

Adobe-PDF-downloadHamilton, Brandon Kyle. MURAC: A unified machine model for heterogeneous computers. PhD Thesis. Department of Electrical Engineering, University of Cape Town, 2015.

 

Abstract:

Heterogeneous computing enables the performance and energy advantages of multiple distinct processing architectures to be efficiently exploited within a single machine. These systems are capable of delivering large performance increases by matching the applications to architectures that are most suited to them. The Multiple Runtime-reconfigurable Architecture Computer (MURAC) model has been proposed to tackle the problems commonly found in the design and usage of these machines. This model presents a system-level approach that creates a clear separation of concerns between the system implementer and the application developer.

The three key concepts that make up the MURAC model are a unified machine model, a unified instruction stream and a unified memory space. A simple programming model built upon these abstractions provides a consistent interface for interacting with the underlying machine to the user application. This programming model simplifies application partitioning between hardware and software and allows the easy integration of different execution models within the single control ow of a mixed-architecture application. The theoretical and practical trade-offs of the proposed model have been explored through the design of several systems. An instruction-accurate system simulator has been developed that supports the simulated execution of mixed-architecture applications.

An embedded System-on-Chip implementation has been used to measure the overhead in hardware resources required to support the model, which was found to be minimal. An implementation of the model within an operating system on a tightly-coupled reconfigurable processor platform has been created. This implementation is used to extend the software scheduler to allow for the full support of mixed-architecture applications in a multitasking environment.

Different scheduling strategies have been tested using this scheduler for mixed-architecture applications. The design and implementation of these systems has shown that a unified abstraction model for heterogeneous computers provides important usability benefits to system and application designers. These benefits are achieved through a consistent view of the multiple different architectures to the operating system and user applications. This allows them to focus on achieving their performance and efficiency goals by gaining the benefits of different execution models during runtime without the complex implementation details of the system-level synchronisation and coordination.