Karthikeyan Bhargavan, Andrew Gordon, Tim Harris, and Peter Toft
Various hardware and software technologies are transforming datacentre management into a programming problem. Network booting and virtualization allow dynamic allocation of hardware resources. Virtual machine mobility helps to optimize hardware utilization, to support maintenance, and to minimize power consumption. Virtual networking isolates mutually distrustful workloads, and allows test to proceed concurrently with production. These technologies reduce the need for human intervention - indeed, virtual machines can even be rented by the hour over the web, eliminating physical configuration altogether. These hardware and software trends support the abstraction that a datacentre (or even a collection of datacentres) is a single programmable computer, made up of many individual servers managed in software. Going by different names – compute fabric, grid, utility computing, autonomous computing, dynamic systems, cloud – this abstraction has arisen in different research communities, albeit with varying emphases. To help identify discussion topics at the workshop we invited participants to contribute short (unrefereed) position papers about their insights into the problems of managing this kind of system. This elicited an exciting response from a wide range of disciplines – covering, amongst other topics, practical experiences working with today’s datacentres, foundational aspects of security and service design, technology trends, and performance modelling.