Good to see you mentioning log structured trees as well. It would be very interesting to understand how they compare to Fractal trees by Tokutek. Look very similar in terms of advantages offered to me. The good thing about them – as I understand they are not patented (at least they are implemented in number of technologies)

I think Tokutek and Infobright have some overlap of simple analytic queries but their focus is very different.

Nice analysis of TokuDB. In regard to Infobright, we are designed for analytics rather than transaction processing. Our technology is a column oriented database based on MySQL, that uses what we call the “knowledge grid”, which eliminates the need to create any indexes or partition data while achieving great query performance. We also sport very high compression (better than 10:1, sometimes as high as 40:1). You can think of the knowledge grid as a metadata layer which describes at a high level the content of the entire database. It is used by our optimizer to eliminate as much data as possible when processing a query and minimize the need for decompression. Since we don’t have indexes, our load speed is one the fastest you’ll find. We have quite a few clients that are successfully using Infobright in production for web log analytics. We are also open-source so I invite you to check it out at http://www.infobright.org.

Cheers –V

In the mean time, her are links to the papers (any difficulty in reading them is certainly the fault of our academic writing rather than of the understandability of the math!):

http://www.cs.sunysb.edu/~bender/pub/sicomp05-BenderDeFa.ps
http://www.cs.sunysb.edu/~bender/pub/FOCS03-co-searching.ps
http://www.cs.sunysb.edu/~bender/pub/locality-full.ps
http://www.cs.sunysb.edu/~bender/pub/BenderHu-TODS07.pdf
http://supertech.csail.mit.edu/cacheObliviousBTree.html

FYI: there’s a totally different index structure called Fractal Trees (see e.g. http://www.pdl.cmu.edu/ftp/Database/fpbtree.ps). This is unrelated to our work.