Cost/Performance in Modern Data Stores – How Data Cashing Systems Succeed

Data in traditional “caching” data systems resides on secondary storage, and is read into main memory only when operated on. This limits system performance. Main memory data stores with data always in main memory are much faster. But this performance comes at a cost. In this paper, we analyze the costs of both in-memory operations and secondary storage operations where data is not “in cache”. We study the performance impact of cache misses on caching system performance. The analysis considers both execution and storage costs. Based on our analysis, we derive cost/performance results for a data caching system [Deuteronomy and its Bw-tree] and a main memory system [MassTree] to understand where each demonstrates the best cost per operation, what is driving the cost differences, and the scale of the differences. This analysis (1) provides insight into why data caching systems continue to dominate the market; (2) points to higher performance that does not rely on simply increasing main memory cache size; and (3) suggests a path to lower costs and hence better cost/performance.

Speaker Details

David Lomet is a principal researcher in and founder of the Database Group at Microsoft Research, Redmond. Earlier, he worked at DEC’s Cambridge Research Lab, at IBM Research Yorktown, and was a professor at Wang Institute. He has a Ph.D. from the University of Pennsylvania. David has worked in machine architecture, programming languages, and distributed systems. His primary work is in database systems and he is one of the inventors of transactions. His most recent work is on the Deuteronomy project, whose transactional key-value store exploits clean architecture and provides high performance. He has authored over 100 papers, including two SIGMOD “best papers”, and holds 60 patents. David is a Fellow of IEEE (and Golden Core Member), ACM, and AAAS. He was awarded the IEEE Meritorious Service Award and SIGMOD Contributions Award for serving as Editor-in-Chief of the IEEE Data Engineering Bulletin since 1992. He served on the IEEE Computer Society Board of Governors, and is currently 1st Vice President and Treasurer. He has been ICDE PC co-chair and conference co-chair, VLDB PC co-chair, and editor of ACM TODS and VLDB Journal. He has served on the VLDB Board and the ICDE Steering Committee, and has been IEEE TCDE Chair.

Date:
Speakers:
David Lomet