Kees Goossens

Formal resource management for real-time multicore applications

Abstract

In practice almost all computer systems are limited by the available resources (compute, memory, energy or power). In a cloud the service quality given to the current set of applications has to be managed. Embedded & edge systems are affected doubly: a higher quality is usually required (e.g. hard deadlines, low jitter) but using fewer resources.

Since average performance is the focus of current hardware and software architectures they require only a limited awareness of resources. In this presentation we will present CompSOC that takes the opposite approach, accounting for every clock cycle and byte in both hardware and software. As a result of this precise resource management, predictable & composable Virtual Execution Platforms (VEP) can be offered to applications. VEPs simplify the verification of application development & performance.

However, deploying multiple VEPs introduces a quality-resource optimisation problem. We will briefly introduce a formal component-based quality-resource model, as implemented in the QRML language and its Pareto-based optimisation.

Much interesting work remains to be done, especially in solving the gap between offline (formal) design and online quality-resource optimisation.

Bio

Kees Goossens has a BSc in computer science from the University of Wales and a PhD from the University of Edinburgh. 

He joined Philips Research in 1995, where he worked on behavioural synthesis for high-throughput video processing, on-chip communication protocols and memory management.

He is currently a full professor at the Eindhoven University of Technology, where his research focuses on composable (virtualised), predictable (real-time), low-power embedded systems, supporting multiple models of computation.

He is the author of 24 patents and published four books, and 175+ articles, with four paper awards. The DATE conference selected his 2003 paper as one of the 30 most influential papers in 10 years.