Big Data: Riak or Cassandra?

Hey guys,

I am a system administrator and was given the incredible well documented task to support "Big Data". In this particular case, the idea is to get an impression of the user behaviour on a quite big Swiss website by the means of aggregating all log data and querying it later for analysis.

Our Java coders are mostly settled to "Let's do cassandra!" while Riak also looks interesting. Having to support "schema changes" and running it in a mission critical environment (i.e. 0% downtime) are some requirements.

We already have mongodb running, but mongodb's sharding makes life hard when you are a sysadmin and think about backups.

So my question to you is:

Given the two alternatives, which one would you choose if it was you to run the infrastructure?