Your CTO is right. I was recently in the position of using Tokyo Cabinet on a site with a lot of traffic. Initially, it was great. It continued to be great for a long while until we hit a point where it started taking up too much memory and 100% cpu causing load to sky rocket. While the technology is great, there was very little that Google found about our particular problem. We eventually wound up resetting our tyrant database and plan on moving completely off of it.

You didn't strace it to see what was going on?

> documentation is spotty

True, but we're working on that (and we're very responsive on IRC and mailing lists).

> not enough big projects are using it

I don't think that's true anymore; Digg, Facebook, Rackspace, Twitter, Mahalo, Ooyala, SimpleGeo, WebEx, and others are using Cassandra: http://n2.nabble.com/Cassandra-users-survey-td4040068.html

Thanks for the link, it will be useful if I try to make a case for cassandra vs redis. Other questions: how robust are the clients (especially for php), and are you aware of any performance/scaling comparisons between cassandra and redis?

the clients are thrift, and the commonly used ones (java, python, ruby, C++, php), are robust, although the thrift devs are slow to apply patches sometimes (all the php users I know have applied http://issues.apache.org/jira/browse/THRIFT-347).

redis is faster per node, but does not scale, there is not much to compare. :)