Abstract

Heterogeneous multi-core architectures (HMAs), such as the Cell processor in the Playstation 3, incorporate different core types on a single CPU . These HMAs have the potential to significantly increase application performance, however, they are notoriously difficult to exploit effectively and thus their use is currently restricted to specialist domains.

This talk will present Hera-JVM, a runtime system aimed at enabling non-specialist programmers to exploit HMAs. Hera-JVM supports the execution of standard multi-threaded Java applications on the disparate core types of the Cell processor. The behaviour of a thread is tracked by the runtime system, using either explicit code annotations or runtime monitoring. This behavioural information is used by Hera-JVM to automatically migrate threads between the Cell’s heterogeneous processing cores, to gain the best performance from the architecture. Hera-JVM has also been ported to a non-uniform memory access (NUMA) based x86 server. On this system it uses a thread’s behaviour characteristics to try and optimise thread placement on the different NUMA nodes, such that inter-thread communication overheads are reduced.

This talk will describe the approach taken to by Hera-JVM for these two heterogeneous systems, as well as presenting experimental results from benchmarks run under Hera-JVM on both systems.

Bio: Ross McIlroy is a Post-Doc researcher in the System’s and Networking group at Microsoft Research Cambridge. His main research interests lie in operating system, runtime system and programming language support for multi-core architectures, particularly the interface between application programs and heterogeneous architectures. He is currently involved in the Barrelfish operating system, a joint collaboration between MSR and ETH -Zurich.

His has recently submitted his PhD thesis at the University of Glasgow. This thesis involved the creation of a Java Virtual Machine for the Cell processor, called Hera-JVM. He completed a MSci in Computing Science in 2005, also at the University of Glasgow. This involved a research project that employed the Xen Virtual Machine Monitor to partition a network router’s resources between multiple competing QoS network flows.