File and Storage System Design

Our current work focuses on designing file systems and their supporting operating system architecture. File systems must evolve as system architectures change around them. We aim to identify these changes and understand how file system architecture should respond. Current work includes:

• Re-evaluating the design of distributed file systems within the context of zero-copy architectures. We are actively participating in the Direct Access File System (DAFS) collaborative effort.

• Developing a generalized logging facility from a journaling core. Using this for strong data protection within storage-area network environments.

• Designing flexible file system interfaces and application-specific file systems.

• How should the performance of file systems be measured?

• How do file systems age and how can this aging be simulated to improve the relevance of benchmarks?

• How do soft updates and journaling differ in performance and semantics? (With researchers at CMU and Kirk McKusick)

• How do the different FFS allocation algorithms compare?

• How do clustering and file system logging compare?

This research is part of the VINO project.

Researchers

• Professor Margo Seltzer
• Salimah Addetia
• Dan Ellard
• Kostas Magoutis
• Lex Stein

Alumni

• Keith A. Smith, Graduated with Ph.D.. Now at Sun Microsystems.

Publications

• Keith Arnold Smith, Workload-Specific File System Benchmarks, A thesis presented by Keith Arnold Smith to The Division of Engineering and Applied Sciences in partial fulfillment of the requirements for the Doctor of Philosophy in the subject of Computer Science, Harvard University, Cambridge, MA, January, 2001, 159 pages.

• Margo I. Seltzer, Gregory R. Ganger, M. Kirk McKusick, Keith A. Smith, Craig A.N. Soules, and Christopher A. Stein, Journaling versus Soft Updates: Asynchronous Meta-data Protection in File Systems Proceedings of the 2000 USENIX Annual Technical Conference, June 2000, San Diego, CA, pages 71-84.

  This paper presents two journaling file system implementations and compares their performance with that of soft updates, all on the same platform within a BSD operating system. Journaling uses an auxiliary log to record meta-data operations while soft updates uses ordered writes to ensure meta-data consistency. Journaling can be run either synchronously or asynchronously. The mode affects the file system semantics. We toggle the semantics of journaling and compare with both the upper-bound of an unrecoverable file system and soft updates to explore the relative costs of semantic properties.
  Benchmarks
  Full Abstract

• Keith A. Smith and Margo I. Seltzer, File System Aging - Increasing the Relevance of File System Benchmarks, Proceedings of the 1997 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, June 1997, Seattle, WA.

  Many file system benchmarks are executed on empty file systems, a state that few users encounter in the real world. To allow researchers to measure file systems in a more realistic setting, this paper presents a technique called file system aging that uses an artificial workload to simulate the effects of many months, or even years, of activity on a test file system. This paper also demonstrates the use of file system aging by using it to evaluate to modifications to the file layout policies of the UNIX fast file system.

• Keith A. Smith and Margo Seltzer, A Comparison of FFS Disk Allocation Algorithms, Proceedings of the 1996 USENIX Technical Conference, January 1996, San Diego, CA.

  This paper compares two different disk allocation policies implemented for the Berkeley Fast File System. A simulated ten month workload is used to fill and fragment test file systems that use the different allocation policies.

• Seltzer, M., Smith, K., Balakrishnan, H., Chang, J., McMains, S., and Padmanabhan, V., File System Logging versus Clustering: A Performance Comparison, Proceedings of the 1995 USENIX Technical Conference, January 1995, New Orleans, LA, pp. 249--264.

  This paper is a comparison of a log-structured file system to the conventional UNIX file system (FFS). The analysis of FFS includes an analysis of the impact of file fragmentation found in the tech. report (see above).

• Keith A. Smith and Margo Seltzer, File Layout and File System Performance, Harvard Computer Science Technical Report TR-35-94.

  This paper presents empirical data characterizing the fragmentation and performance of FFS file systems that range from several months to several years in age. [Overview].