scientific benchmarking

[dominictarr]

I'm just putting this here for future reference,

https://github.com/brianfrankcooper/YCSB

It's all such/very/much/wow.java etc, but the concepts could probably be implemented as 300 lines of javascript.

The basic idea is fairly simple, it generates random benchmarking workloads,
that are more realistic than what we have right now.

It can set a proportion of reads and writes, and writes are distinguished between updates or inserts. Also, the distribution can be uniform or zipfian (the graph on the power law article is more informative)

The most important part is probably here: https://github.com/brianfrankcooper/YCSB/tree/master/workloads

That is the different models they benchmark with, so I'm thinking that the only tricky part here is the zipfian distribution.