Remember the Riak in the Joyent cloud benchmark? There’s a post providing many more details about the tests run:
The goal of the study was to demonstrate a baseline for users to understand Riak’s performance, stability, predictability, and linear scalability. The systems were not tuned for optimal performance. Instead, we chose to take standard 4 GB Riak Smartmachines and demonstrate throughput and latency for various access patterns and object sizes.
The conclusions is what made me say it got atypical (in a good sense) results:
Our benchmark tests bring us to the following conclusions:
- Riak behaves predictably under high loads – depending on system resources, Riak exhibits either predictable, steady-state throughput with low errors or degrades gracefully with low errors.
- Riak demonstrates stability under high loads – very few errors, no node failures under load, and behavior in line with expectations.
- Riak demonstrates linear scalability – adding or removing capacity adds or subtracts a predictable amount of capacity from the cluster.
Original title and link: Riak SmartMachine Benchmark: The Technical Details (NoSQL databases © myNoSQL)