|
New England Database
Society sponsored by Sun Microsystems |
| |
NEDS |
Coordinating the Database and Storage Tiers
Ken Salem
University of Waterloo
Friday, December 1,
2006, 4:00 PM
Volen 101,
Brandeis University
(preceded by a wine and cheese reception at 3:00 pm)
Abstract:
Database management systems have a simplistic view of the underlying storage systems that serve them. Conversely, storage systems have a limited understanding of the database systems and other clients that they are expected to serve. This talk will cover some ongoing work on two problems that arise because of this knowledge gap.
The first problem is two-tier cache management, where the first tier cache is in the database system and the second tier is in the storage system. Naive management of these two caches can effectively waste the capacity of one of them, hurting performance. I will describe a technique for alleviating this problem by providing the storage tier with additional application knowledge via hints. The storage tier exploits these hints to more effectively manage its cache.
The second problem is multi-tier physical design. Storage systems, like database management systems, must be properly configured if they are to perform well. Although the database physical design and storage configuration problems are closely related, the database and storage tiers are often separately administered. This separation makes it difficult to achieve a good end-to-end design. I will present a technique for storage workload estimation that is intended to help address this problem. Our storage workload estimator translates a given database workload and database physical design into a characterization of the storage workload that will be generated by the database system. Such a characterization can then be used by a storage administrator to guide storage configuration.
Speaker Bio:
Ken Salem is an associate professor in the David R. Cheriton School of Computer
Science at the University of Waterloo. He received his B.Sc. in electrical
engineering from Carnegie Mellon University, and his Ph.D. in computer science
from Princeton University. Before coming to Waterloo, he spent several years as
a member of the faculty at the University of Maryland, College Park, and he has
spent sabbatical leaves at IBM's Almaden Research Center and at HP
Laboratories. His research interests span a variety of topics related to
database management systems.
Maintained by Dina Goldin dqg AT cs.brown.edu