Abstract:
Many multiprocessor applications have performance that depends on the efficiency of operating system services. Unfortunately, current operating systems often have performance that scales poorly with the number of cores, because they implement their kernel abstractions using data structures modified by all cores. On processors with many cores this design may cause the kernel to be a bottleneck due to the costs of inter-core interactions and data movement. This talk proposes several new kernel abstractions (e.g., address trees, kernel cores, and shares), which can improve kernel scalability, and a design principle (applications should control sharing), which has guided the design of these new abstractions.

Joint work with: S. Boyd-Wickizer, F. Kaashoek, R. Morris, A. Pesterev, Y. Zhang (MIT) and H. Chen, R. Chen, Y. Mao, L. Stein, M. Wu, Y. Dai, Z. Zhang (MSRA)

Biography:
M. Frans Kaashoek is a full professor in MIT's EECS department and a member of Computer Science and Artificial Intelligence Laboratory, where he coleads the parallel and distributed operating systems group (http://www.pdos.csail.mit.edu/). He received a PhD (1992) from the Vrije Universiteit (Amsterdam, The Netherlands) for his work on group communication in the Amoeba distributed operating system, under the supervision of A.S. Tanenbaum. Frans's principal field of interest is designing and building computer systems. His past work includes the exokernel operating system, the Click modular router, the RON overlay, the self-certifying file system, the Chord distributed hash table, and the Asbestos secure operating system. Frans is a member of the National Academy of Engineering and the recipient of several awards, including the inaugural ACM SIGOPS Mark Weiser award for demonstrating creativity and innovation in operating systems research.