Jeremy Singer, Chris Kirkham and Ian Watson
School of Computer Science, University of Manchester, UK

Virtualization is ubiqitous, with the global availability of the Java Virtual Machine and other similar virtual machine platforms. Higher-order virtualization involves building a stack of virtual machine layers. This provides obvious advantages such as: flexibility; separation of concerns; reuse of existing functionality; support for legacy platforms. However, the benefits of higher-order virtualization come at a price in terms of efficiency. This paper examines different techniques to increase the efficiency of higher-order virtualization on chip multiprocessor architectures. These techniques embrace hardware, software and virtual machine interaction. Some techniques (such as just-in-time compilation) are employed in existing virtual machine systems. Other techniques (such as hints-based speculative parallelism) have not been previously considered. We examine how to use these perfomance-enhancing techniques in the context of stacked virtual machine layers.

Back to the Workshop Program