Markets and Clouds: Adaptive and Resilient Computational Resource Allocation Inspired by Economics

Since the earliest days of computers, people have sought to apply them to the solving of large and complex problems. Indeed, computers’ ability to solve large problems have brought benefits to humanity in fields as wide ranging as chess playing [1] and protein folding [2], amongst many others. However, key to the continued ability to apply computers to these kinds of problems is finding ways to enable them to scale massively, while remaining accessible to those who might use them. For example, it could be argued that the requirement either to own a supercomputer such as Deep Blue or have the funds and specialist knowledge to build a distributed platform such as that used by the Folding@Home project reduces accessibility. Grid computing is one technology which attempts to address this. By providing a standard way to access computing power on tap, a grid platform allows users to run very large generic programs, distributed over many computational nodes [3]. Related technologies such as cloud computing [4] enable a similar standard means of access to potentially unbounded scalable computing, while service oriented architectures [5] provide a framework for distributed computational resources to be componentised and packaged up, such that distributed applications may be constructed from loosely coupled components. As platforms grow, localised failures become more likely, and systems can often no longer be assumed to be of a static nature, as its component nodes can be added and removed during run time. Therefore, key characteristics for resource allocation mechanisms to possess are that they are resilient to failures and also able to adapt in order to obtain high performance, taking advantage of system changes during run time. Given such a range of approaches to scaling up computational capabilities, it is not surprising that computational resource allocation in such systems is not a single well defined problem. Instead, it is perhaps best described as a