I’m benchmarking Aerospike cluster via Yahoo Benchmark. Slightly modified sync client is used (modification – array of nodes’ hosts instead of one node in the client constructor).
Pure read throughput doesn’t depend on amount of nodes in cluster. I assume, it happens because one node has huge queue during benchmark execution - ~470 items, whereas other nodes have only 0-2 items.
Replication factor 2.
If you need additional configuration/metrics, let me know.