CAREER: Quickstep: An Architecture-Conscious Main Memory DBMS

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

With the trend toward denser and cheaper memory, database servers in the near future will have large main memory configurations and a significant number of datasets will reside in main memory. In this configuration, the performance of the database system is limited by how fast the memory subsystem can feed instructions and data to the processor. Traditional data storage and query processing techniques are optimized to minimize access to disks and do not effectively utilize the processor caches, resulting in poor query performance. The goal of this research project is to investigate, and experimentally evaluate, data storage and query processing techniques that can effectively utilize the processor caches. The approach consists of novel data storage techniques that can incrementally and transparently adapt the layout of data to better utilize the processor caches, cache-conscious indexing techniques for both traditional alphanumeric data and richer data types such as spatial data, and cache-conscious query evaluation and scheduling techniques. A prototype database system, called Quickstep, is being developed to experimentally evaluate the proposed techniques. The results of this project will produce techniques for efficient query evaluation in environments where the data is mostly resident in main-memory, with possible application in high-performance database servers powering Internet sites and future network routers that will require query processing for providing application-specific routing services. The education component of this project aims to introduce material in the graduate-level database course to expose students to the emerging confluence of databases and architectural technologies. The systems-building effort of this project provides an environment for training both undergraduate and graduate students in database implementation and performance evaluation techniques.