Comparing Hosted Database Performance
compose.io
compose.io
Also to provide some article-related input, in some NoSQL style databases latency makes a very surprising difference in speed. Redis is especially affected. When network latency is introduced on even the 50ms level GET/SET operations drop to double digit (vs. 15k+/s in normal operation even on small servers)
The point of this is to draw attention to the general shape of the behavior. E.g. spikes, latency, increase/decrease over time. The actual timings don't seem that useful to me, since the tests are just built to strain some part of the infrastructure.
The NoSQL point is interesting. I haven't used Redis but I've seen that the behavior you're describing in Hazelcast, which is the same type of system. It would be interesting to do these types of tests while adding additional servers to that type of system.