Colla Rsch:CRD: Community Resource Development: An Open Source
Project Proposed: This collaborative project, developing an extensible open source Virtual Machine Monitor (VMM) for modern architectures (those that support the Intel VT or AMD Pacifica virtualization extensions), aims to maintain a small codebase size while supporting extensive extensibility. The approach combines compile-time composition of major modules to configure the VMM with run-time extensibility akin to a microkernel. The project is expected to produce a fundamental tool for research in systems and architecture, an open source extensible virtual machine monitor. Expecting to meet the needs within the high-end computing community identified by the FAST-OS effort, the VMM will be used as a shared community resource.
Broader Impacts: The infrastructure will be useful in research and education and production context. It will be free and available to all in source code form. Underrepresented groups are impacted through Northwestern's AGEP program and through U NM, a minority serving university.
The primary focus of this community resource development effort has been to design, implement, and enhance software to support U.S. and partner research in virtualization. Virtualization, which underlies more widely known technologies such as cloud computing, involves operating systems software that gives a single physical machine the appearance of acting like multiple virtual machines. To the software running inside it, a virtual machine is indistinguishable from a physical machine. In this project, we developed such systems software, which we named the Palacios virtual machine monitor. Palacios has seen four releases over the course of the project, and currently comprises over 150,000 lines of C and assembly code that was implemented from scratch at Northwestern and the University of New Mexico. The software is provided under an open-source BSD license, making it very easy for other researchers, and commercial entities, to use, extend, or leverage it. Anyone can access the software from our project web site, v3vee.org. It has been downloaded about 50,000 times so far. We ourselves have leveraged Palacios to conduct research into virtualization for high performance computing. We developed techniques that allow us to virtualize a large scale supercomputer, such as Red Storm at Sandia National Labs, with minimal overhead, even at scales of up to 6000 nodes. We also developed techniques to allow for virtualizing networking at the speeds needed in high end computing systems. Our results suggest that virtualization of supercomputers is very feasible, and that doing so might allow a blurring of the line between supercomputing and cloud computing. This could result in the wider use of high end supercomputing, as well as permitting scientific computing needs to be met with a broader set of computers. The project has also contributed to education and training. A new course at Northwestern, in operating systems design and implementation, has been developed around Palacios, and it has trained 60 students thus far. Several Ph.D. students have been produced by the project. Numerous undergraduates, including those from under-represented groups, have been involved and supported, with three undergraduates becoming co-authors on research publications. The project has continuing broader impact. We and our collaborators at the University of New Mexico, the University of Pittsburgh, Sandia National Labs, and Oak Ridge National Labs are continuing the development of Palacios and research on virtualization in hgih performance computing as part of the Department of Energy's X-Stack program that is investigating systems software for the next generation of the nation's supercomputers.
