CSR-EHS: Virtual-Machine Techniques for Resource-Constrained Devices: Reconciling Reliability With Reusability and Low Development Costs in the Embedded Systems Space

University Of California-Irvine, Irvine CA

Investigators

Abstract

Virtual machines (VMs) are widely-used computer-based execution frameworks that isolate an application from the underlying hardware and system software platform and implement programming language run-time semantics.. This project focuses on efficient execution of virtual machine code on resource-constrained devices by way of trace-driven dynamic compilation. Many optimization algorithms that are expensive to perform on graphs can be performed in linear time on a trace. Furthermore, a trace-directed compiler can in-line method calls in a straightforward manner, significantly reducing the modularization overhead of object-oriented programming where it is most critical. This approach thereby paves the way for good software engineering practice and component-oriented programming in embedded software without the accompanying performance penalties that are often associated with these techniques. The project is founded on a novel use of Static Single Assignment form and preliminary results point the way to a completely new way of designing embedded hybrid VMs. The main focus is on systematically exploring and further developing this approach, initially targeting the Java Virtual Machine (JVM) platform. A second focus is on developing a scalable VM layer that is optimally situated to support a wide variety of embedded software, but that can yet be efficiently mapped onto current and future embedded processor platforms. Embedded systems present an application of the 'write once, run forever after' computing paradigm that has not yet reached its true potential. The project is expected to lead to significant improvements in the way embedded software is developed. It will enable reuse of embedded software where no such reuse is currently possible and simplify the transition to increasingly evolved embedded processors. This is expected to have impact in significantly lower cost and significantly increased reliability of embedded software.