TOCS Guest Editorial for SOSP special issue

TOCS Guest Editorial for SOSP special issue

This year I was very privileged. I read eighty-four fine systems papers representing the best work of the past two years. I led a team of thirteen outstanding individuals who selected twenty-two of those papers for full conference presentation and eight papers for conference posters. But then I did not feel so lucky. We had to recommend just a few of these papers for the Transactions on Computer Systems. Somehow we managed to choose, and after further rigorous refereeing by TOCS, you are now lucky enough have them here before you. They are the best of the best in diverse and important areas of operating systems.

The Measured Performance of Personal Computer Operating Systems is a milestone in objective comparison of full systems. Its methods are a new benchmark for SOSP, its results are exciting for both the theory and practice of operating system design, and, unusually for SOSP, it is interesting reading for anyone with a computer science background and interest in the performance of the dominant O.S. platforms.

It is the trend in operating systems work of the 1990s to find principled ways to remove performance and functionality barriers introduced by the abstractions of the 1970s and 80s. Serverless Network File Systems is leading edge work that tightly integrates two classic areas of O.S. work: file systems and distributed systems. To be successful, such an integration needs great attention to detail in theoretical underpinnings, attention to past lessons, and excellent implementation and measurement. This paper achieves that high standard, and well repays close reading.

As computers become more ubiquitous their reliability becomes more important. Hypervisor-based Fault-tolerance is a detailed study of how to implement classical replication based fault tolerance on existing hardware and operating systems. This work is part of a trend, including general purpose media processors and sandbagging-based memory protection, of doing in software what traditionally has required special hardware. I expect this outstanding paper will seed a flowering of much more research in software-based fault tolerance.

One of the most successful file system concepts of the past fifteen years is that of the Redundant Array of Independent Disks, or RAID. There are several excellent RAID products, and yet there are still many unsolved problems, particularly regarding managing the complexity of a complete RAID-based storage subsystem. The HP AutoRAID Hierarchical Storage System significantly advances the art of the RAID by introducing a new method for achieving high-performance, scaleable, and portable RAID architectures.

Science grows through an open exchange of information by which scientists build upon the work of others. I believe that exchanging system source code is as important to operating systems research as the human genome project is to biology. I am very pleased that eight of the twenty-two accepted SOSP papers are making the source code for their systems available to other researchers. Of the four papers in this issue of TOCS, source code is available for three: The Measured Performance of Personal Computer Operating Systems, Serverless Network File Systems, and The HP AutoRAID Hierarchical Storage System. The SOSP CDROM, produced by Andrew Birrell, contains this information as well as electronic versions of all the papers.

I thank the SOSP program committee for its energy, time, and wisdom. I especially thank Frans Kaashoek who excellently led the four papers in this issue through the TOCS refereeing and acceptance process. Finally, I thank the entire operating systems community for their fine body of work that created the opportunity for the four outstanding papers presented below, of which we can therefore all be proud.

Mark Weiser (
last updated November 30, 1995