kms

Benchmarking Large Riak Data Types, continued

December 2, 2014

Yesterday’s post could have included a bit more information. Here’s that information.

My environment

I’m on Mac OS X 10.9.5, and I ran the benchmark against a single Riak 2.0.2 node running locally, installed from homebrew, which does a binary install.

Other graphs

Some graphs I didn’t include were a comparison of byte size and number of items, Set read times for both the key-value and data-type APIs, and Map read times for the key-value API. Here they are:

Set byte size and number of items
Set byte size and number of items
Map byte size and number of items
Map byte size and number of items
Set reads (key-value API)
Set reads (key-value API)
Set reads (data-type API)
Set reads (data-type API)
Map reads (key-value API)
Map reads (key-value API)

Raw data

Here you go!

A potential fix

Russell Brown is working on a feature branch of riak_dt which may have a fix for some of this. I’m running the benchmark against a locally-compiled Riak right now, and then I’ll build it with his potential fix and compare the benchmark results.

Update: I ran the benchmark against the potential fix with good results.