Computer memory systems recently have become stunningly complex. Performance is no longer simply a matter of spatial and temporal locality, but of matching the memory access pattern to the set of patterns for which the processor was designed. Not just for memory accesses, but in general, modern processors hate surprises! Thus, we think in terms of minimizing entropy of a memory access pattern as compared to the access patterns for which the processor was architected....
Something else has happened too: the penalty for a bad reference pattern has jumped to hundreds of clock cycles per reference. Even spending dozens of clock cycles to improve each reference can yield significant speedup.
This page will present the techniques we have developed using compression technology to reduce memory access entropy (not necessarily decreasing memory footprint). Foremost of the new technologies is something we call compressive hashing....
Our first paper on this topic appeared at LCPC 2002; a preprint of it is available as this PDF.
The only thing set in stone is our name.