[Jonathan Appavoo] Cloud Software Systems

3:00 PM Wednesday, April 10th, 2013

EbbRT – A Modern Approach to Systems Software for Cloud Applications

Jonathan Appavoo
Computer Science

Abstract: Cloud computing has and continues to transform the kind of computational platforms that an individual can access. Slowly but surely it will be possible to economically access shared public supercomputer systems. Today’s cloud computing applications, however, are built on top of legacy operating systems (OS’s) that are designed for small-scale shared memory multiprocessors. While these OS’s make it possible to construct applications from a rich body of existing middleware and libraries they are in some sense out of sync with the integrated nature of supercomputer systems. These OS are also not the most appropriate building block for Cloud applications. Given the pay-as-you-go model Cloud applications ideally would be able to quickly and dynamically grow and shrink their footprint across the resources a platform. Legacy OS’s such as Linux and Windows are not designed for this degree of elasticity or scale. Nor do such operating systems fundamentally incorporate primitives for developing and executing applications that span multiple nodes and exploit the available high-performance communications facilities. On the other hand while the runtime/OS model of supercomputers, such as the IBM Blue Gene family, adopt a dedicated light-weight, kernel model they do not address the needs of elasticity nor do they provide a framework for general purpose service oriented applications. In this talk we describe our work on a new system software architecture and runtime that targets the challenges associated with developing software that can efficiently leveraging such systems for a broad spectrum of applications and use cases. We describe this system software effort in the context of the lessons from our past systems research and our on going effort to develop an “Interactive High-Performance Computing” service for high-speed elastic 3D fetal neuro-imaging reconstruction.

Bio: Jonathan Appavoo joined the Department of Computer Science in 2009. Before coming to Boston University, Jonathan was a member of the research staff at IBM T.J. Watson Research Center in New York. Jonathan received his Ph.D in Computer Science from the University of Toronto in 2005. His work focused on scalable systems software for large-scale, general purpose multi-processors. Specifically, he worked on an object model for systems software construction which permits and encourages fine grain control of sharing and attendant communication costs. He first pursued this work in the University of Toronto, Tornado operating system and then in the IBM K42 operating system. In 2007, along with his colleagues Volkmar Uhlig and Amos Waterland, he established Project Kittyhawk to explore conjectures about Global Scale computers and computation. The project took a holistic approach to the provisioning online computational capacity influenced not only by technical aspects but also socio-economic aspects. The goal was to explore computation as a more basic commodity for the construction of various digital products and services including cloud computing itself. His current research interest are two fold 1) Exploring new system software for developing and deploying scalable and elastic applications on future data-center scale systems and 2) Pursing a new kind of “Programmable Smart Machine” — hybrid computing systems that behave as programmed but transparently learn and automatically improve their operation.