Smallbase is a main-memory relational DBMS developed at HP Labs. The performance of the system is an order of magnitude faster than that of disk-based DBMSs in both throughput and response time. In this presentation, I will relate key factors that have contributed to the high performance of Smallbase. I will also relate what we have learned about user requirements in high-performance data management. Most importantly, while a common belief is that OLTP performance is as good as it needs to be, we have evidence to the contrary from users in numerous application domains.
The advent of 64-bit computers and the resumption of drop in RAM prices should be incentives to researchers to rekindle the interest in main memory data management. Of particular interest to users are hybrid solutions that involve both main-memory data management and disk-based data management. I will discuss some results we have in this area and will outline other research opportunities.
Smallbase is joint work with Sheralyn Listgarten, Kurt Shoens, and Kevin Wilkinson from HP Labs, and Oded Shmueli from Technion.