|
The ADvanced Systems Laboratory (ADSL)
|
||||||||||||||||
|
Sun Sponsored ResearchThis page describes the research performed by the ADSL group that has been partially sponsored by Sun Microsystems. The donation our group has received consists of a cluster of 24 Sun Ultra-20 workstations. These machines provide our group with a premier testbed for experimental computer systems research. OverviewOur current work is in the area reliability. In particular, we are interested in analyzing existing systems and developing new approaches. Thus far, the Sun cluster has played a crucial role in enabling us to experiment with both Solaris and Linux file systems; further, it opens up new educational possibilities in our advanced software systems courses. Research OutcomesOur most recent paper utilizes the cluster to build a new environment for reliable I/O. Entitled "Increasing File System Reliability with I/O Shepherding" (SOSP '07), the abstract is as follows: We introduce I/O shepherding to increase the reliability of file systems in the face of storage system failures. The I/O shepherd is a layer beneath the file system; by interposing on each I/O, the shepherd takes responsibility for its execution as specified by a configurable reliability policy. We develop a framework for flexibly describing these policies, and show how a variety of interesting and relevant policies can be assembled from basic primitives. We integrate I/O shepherding with Linux ext3; the resulting file system, CrookFS, incorporates I/O shepherding through a set of changes to the consistency management subsystem, layout engine, disk scheduler, and buffer cache. We demonstrate how a broad class of policies can be easily and correctly specified, and also show that shepherding is properly integrated into the system, adding 5% overhead in the worst case to the I/O path. More details can be found in the full paper here. One might note the acknowledgement section, which lists Sun (in addition to NSF and Network Appliance) as one of the main sponsors of this work. Of course, beyond this current effort, we have many plans for future directions, all of which is enabled by the donation. From a research perspective, we are currently building infrastructure to analyze the performance and reliability of ZFS, Sun's new and exciting entry in the file system space. We hope to build expertise in ZFS internals and machinery, and eventually to build contacts between our group and the file systems group within Sun. We believe this avenue of research will prove fruitful in the years to come. Class workFrom an education perspective, we are currently using the cluster as the main computational testbed for the advanced graduate operating systems course (CS 736). By doing course projects on these machines, a new generation of Masters and Ph.D. students will become familiar with Sun hardware and software technology. We plan to extend this use to other classes as well. Our first foray will be the inclusion of the Sun resources in CS 736, the graduate operating systems class. Eventually, we hope to make such machines available at the undergraduate level as well, but we cannot currently as there are not enough machine resources available to support an entire undergraduate class. |
||||||||||||||||