NeuroCloud: Developing a hybrid cloud architecture for neuroscience research

The goal of this project is to develop and deploy a flexible computational environment that can be configured to accommodate researchers' needs by optimising the use of institutionally held resources, while at the same time dynamically utilising commercially hosted resources in response to surges in demand. This project will provide leading engineering and technical solutions and, importantly, practical demonstrations of an efficient, economical and flexible environment for the delivery of computational (and data) resources within Higher Education (HE), and, between HE and commercial providers. We aim to develop an environment to better manage institutionally held research-computing resources by making them available to users as part of a hybrid Cloud, which can be accessed internally by researchers and also externally by collaborators. In addition to providing access to locally held resources, the environment will also serve as an interface to commercially available cloud computing resources, where computational power and storage can be access on demand, in return for payment. A key aim is to create a user experience so seamless that researchers may not be aware which resources - internal or external, CPU or GPU, are being used. This project will help institutions to manage their computational resources in a cost-effective way. This approach has an additional benefit in that it would use less electricity than conventional computational approaches.

This project is a direct extension of our other EPSRC projects and stems from user community engagement with computer scientists and engineers. EPSRC support has provided new opportunities for cross disciplinary research and partnerships with industry. However, in many cases access to adequate computational resources flexible enough to respond to the specific needs of a given project has been a stumbling point. The types of projects that have presented themselves to us involve the creation of high fidelity large scale models of biological processes and validating them against corresponding data sets. Such models are computationally very expensive and the computations tend to scale exponentially with the scale of the model. The data sets against which they are compared also require large processing capacity due to their very highly dimensional nature. These projects arise from both purely internal research led needs as well as from commercial led collaborations. In either case a common issue arises; to purchase or build a new computational cluster devoted to the needs of the project or to compromise some aspect of the project in order to fit the computational modelling and analytic design to rigid institutionally or commercially available computational resources. The development and deployment of a computational environment that can be flexibly configured to accommodate to almost any computational need by optimizing the institutionally held resource and engaging effectively with commercially available cloud configurations will have a significant impact on the management of HE research computing resources and will open new opportunities for collaboration with industrial and commercial partners.