Application-Agnostic, Distributed-Aware Cloud Platforms

Timothy Wood, George Washington University, PI

This project is supported by the National Science Foundation's CAREER Grant #1253575.

Cloud Computing has radically changed how businesses run their applications by allowing a huge number of computers to be economically shared by many different users. The applications running inside these cloud data centers are growing in size and complexity. Even a relatively straightforward web application is likely to be composed of multiple interacting service components such as a web server, a database, and a data cache. The result is a complicated distributed application that may exhibit performance bottlenecks or consistency requirements between components. Unfortunately, existing resource management and reliability tools consider these components individually, and are often unaware of the important relations between them.

This work is predicated on the belief that future data centers must be application-agnostic, yet distributed-computing-aware. Cloud data centers rely on virtualization to partition servers into isolated components, but there are benefits and drawbacks of sending information across the virtualization abstraction layer. This project has explored these trade-offs in the context of memory management (VEE 14), storage (SC 15, IC2E 16) and scheduling (CCGrid 14), with a focus on distributed cloud applications. We have also explored how network function virtualization offers a new opportunity to customize network management for cloud applications in a transparent way (NSDI 14, ICAC 2016 Best Paper). The work has resulted in several open source extensions to popular cloud software such as Xen and Hadoop, twenty publications, and a best paper award. A diverse body of students, including four PhD students, six undergraduate students, three MS students, and three high school students, were involved with the research projects.

PhD Students:

MS/Undergraduate Students: Neel Shah (BS '17, VMware), Phil Lopreiato (BS '17, Facebook), Harpreet Singh (MS '17, Stony Brook), Shaohua Duan (MS '16, Rutgers), Chenghu He (MS '16, EMC), Ben Carleton (BS '16), Abigail Shriver (BS '17), Lucas Chaufournier (BS '15, UMass), Rian Shambaugh (BS '15, UMass).

Publications:

Open Source Releases: