Predicting the non-functional properties of a Ceph cluster can be quite difficult. There are many inputs in the hardware setup and software configuration that affect the resulting availability, reliability and performance (latency and throughput at nominal levels and during degraded and rebuild times).

I want to understand the interrelationship of these parameters better and build a (perhaps interactive) model that allows us to predict the result without the need to build a ten thousand node cluster.

I am not yet sure which modelling toolkit to use. Could be ruby, python, Perl or even Wolfram SystemModeler.