Benchmark the filter as a 'server'

[lemire]

Instead of issuing a single query against the filter, we should issue several.

[oschonrock]

I have now integrated my binfuse wrapper of binary_fuse(8|16)_filters into my pawned password project

I put together some top-level metrics of "storage" vs "false +ve rate" vs "query performance"

The performance figures for binfuse are off the scale really, and if there is enough RAM to cache the mmap, then actually that 100,000queries/s is far too small. It's actually that the apache bench is topping out at 100% CPU, and I can't measure it, without using more complex instrumentation. The load is all in generating and handling the http requests, not in running the binfuse->contain() queries.

I have run bulk queries in my binfuse benchmark without the overhead of a client/server/protocol and those were ~50ns, ie ~20million queries/s

[lemire]

!!!