EXPERIENCE WITH GRAY-BOX SYSTEMS

Modifying an operating system is a difficult, costly, and often impractical endeavor. Thus, there are tremendous benefits to being
able to add new functionality to a system without requiring any changes to the underlying OS. Our initial research has shown that a
surprisingly large class of "oS-like" services can be implemented without changing the OS, in particular those services that expose new information to applications. The key to this approach is treating the OS as a "gray box" , in which developers are allowed (and even encouraged) to make assumptions about the OS beyond that specified by the interface. In this talk, I will describe our initial experience with a range of gray-box case studies, focusing on two different gray-box techniques for one particular OS-like service: exposing the contents of the OS buffer cache to I/O-intensive applications, such as web servers. The first technique probes the file cache by reading a single byte from the files of interest; probes which return "quickly" are assumed to be in the file cache and imply that neighboring blocks are in the file cache as well. The second technique simulates the contents of file cache by combining observations of which files are accessed with detailed knowledge of the replacement algorithm; "fingerprinting" techniques are used to automatically determine the replacement algorithm off-line. In the talk, I will discuss the applicability of different graybox technique depending upon the environment as well as how the gray-box approach can be applied more broadly.

Andrea Arpaci-Dusseau is an Assistant Professor in Computer Science at the University of Wisconsin at Madison. Professor Arpaci-Dusseau's research interests are in operating systems and distributed systems. She is currently focused on extending the functionality of existing operating systems through gray-box techniques. She also helped start the Wisconsin Network Disks (WiND) project, studying manageable, heterogeneous Network Attached Storage. Andrea obtained her B.S. degree in Computer Engineering from Carnegie Mellon Univeristy and her M.S. and Ph.D. degrees in Computer Science from U.C. Berkeley.

As a graduate student, she was part of the Network of Workstations (NOW) project and developed implicit coscheduling. She also
likes to periodically help win world records for disk-to-disk sorting.