######################################## #Written by David Tam, 2000. # #davidkftam@netscape.net Copyright 2000# ######################################## David Tam Tuesday, May 2, 2000. Thesis Proposal =============== My research interest is in the area of operating systems. Professor Michael Stumm and his research group are currently collaborating with IBM to develop the K42 operating system (OS). This research OS is being developed to address performance and scalability issues of system software. I am interested in investigating how K42 and its current hardware platform affect file system design. Specifically, I intend to port the Hurricane File System (HFS) to K42 and to study its performance. HFS was designed for a previous generation shared-memory multiprocessor and OS, and consequently contains a valuable framework to build upon. It was originally developed by Krieger [1,2] for the Hurricane OS and Hector multiprocessor. My approach will be experimental in nature, rather than theoretical. Through performance measurement and analysis, I will optimize HFS for K42. Some important issues I will address are (1) applicability of the HFS design, (2) workload selection, (3) component redesign, and (4) HFS design assessment. Performance measurement and analysis will determine if HFS is making efficient and intelligent use of the OS and hardware facilities. An important issue I will address is in finding a suitable workload for evaluation. Possible workloads include various macro benchmarks such as scientific and engineering code, database management systems, transaction processing, client/server, interactive, and streaming multimedia workloads. HFS will be tuned to retain its scalability and high I/O performance properties. Tuning will require redesigning and/or re-implementing various components of HFS. Perhaps modifications to the design of both HFS and K42 will be necessary. K42 and its hardware platform provide various levels of supporting infrastructure and services that may play a significant role in the success of HFS. For instance, in the Hurricane OS and Hector multiprocessor, the availability of an extremely efficient inter-process communication facility (PPC: protected procedure call) plays a major role in minimizing overhead in HFS. The optimization process will lead to an understanding of how K42 and its hardware facilities affect the design, implementation, and performance of file systems. The optimization process will also verify the flexibility, scalability, and robustness of HFS and K42. The current hardware facilities support larger scale multiprocessors but exhibit larger disparities between CPU and memory speeds than was available when HFS was initially conceived. Such a platform will enable stress testing of both HFS and K42. To assess the design choices made in K42 and HFS, the overhead contributed by the hardware, OS, and file system will be examined. For instance, K42 overhead may be due to OS resource contention among independent workloads that are executing simultaneously. In such a situation, the design of K42 will be a major determinant of performance and scalability of a file system. As a part of the analysis, bounds on file system performance will be determined. Actual and potential bottlenecks will be examined along with their dynamics. Understanding the dynamics of bottlenecks may influence future design choices in the operating system and file system. REFERENCES ---------- [1] O. Krieger, "HFS: A Flexible File System For Shared-Memory Multiprocessors," Ph.D. Thesis, Department of Electrical and Computer Engineering, University of Toronto, October 1994, ftp://ftp.cs.toronto.edu/pub/parallel/Okrieg_PhD.ps.Z. [2] O. Krieger and M. Stumm, "HFS: A Performance-Oriented Flexible File System Based on Building-Block Compositions," ACM Transactions on Computer Systems, Vol. 15, No. 3, August 1997, pp. 286-321, http://www.eecg.toronto.edu/~okrieg/papers/hfs_tocs.ps.gz. Student, Supervisor, ______________________ ______________________ David Tam Michael Stumm