The impact of the AlphaSort paper on me was primarily the realization that cache behavior was particularly important for the performance of data intensive operations. I was particularly impressed by the clear way that cache behavior was controlled, and by the high cache hit ratios obtained. I could foresee a time when for many reasonable applications, the entire database could fit into main memory. In that context, cache behavior would be the critical performance factor, since the gap between processor speed and main memory speeds was widening (about 2 orders of magnitude over the last 12 years). This observation would hold not just for sorting, but for all database operations. Cache performance issues are now a central theme in my new main-memory database project at Columbia.
a service of Schloss Dagstuhl - Leibniz Center for Informatics