This project will create an experimental infrastructure for studying cloud computing at Stony Brook University. Cloud computing is the predominant approach for organizing and managing servers for online services such as Amazon, Google, Facebook, and Twitter. Although cloud computing providers are rapidly expanding, very little is clearly understood in the research literature about how best to build and use a cloud. This disparity arises primarily from the extreme secrecy in which cloud computing operators conduct their business, limiting the dissemination of such knowledge to retain competitive advantage. This project will advance the principled study of cloud computing hardware, software, and applications, by giving researchers the ability to directly experiment with a complete cloud infrastructure.
The funding for this project enables the purchase of a small computing cloud to be hosted in the Computer Science building at Stony Brook University, configured to mimic a state-of-the-art datacenter. Users of commercial clouds cannot study many aspects of cloud system design. This project provides researchers with complete physical access to the cloud infrastructure, enabling measurement of the physical characteristics such as power consumption, temperature, and air flow; access to the entire software stack will enable experimentation on cloud resource management and hypervisor implementation; unrestricted access to the cloud resources will enable rapid flexible benchmark and application development and integration. Moreover, direct access to the complete cloud environment, including the locally-developed applications and management infrastructure, will provide researchers with full introspection that enables end-to-end studies of cloud behavior.
This proposal brings together a group of PIs who will conduct cloud-related research with a new level of empirical rigor. Because researchers and educators at Stony Brook have a long list of anticipated projects, the infrastructure should immediately garner heavy use. Among the first projects will be a characterization study of virtualization on the scaling efficiency of cloud applications, which will identify the pressing research opportunities in cloud architecture. Another will begin at-scale measurement and testing of lightweight virtualization techniques. This infrastructure will be leveraged to develop and deploy a cloud-based Accessibility as a Service model for delivering accessible digital content to people with disabilities. This project brings together researchers working at all layers of the cloud computing stack---creating the potential for crosscutting insights from study in a realistic deployment with an active user community.
