Queries here are measured from the application side. And in our case, those are simple key-value operations such as get(key), update(key, value), write(key, value) and delete(key). Aerospike is known to optimize those into fewer block operation as it can.
Yes. Low latency lookups are q requirement. Saying that, even double the latency we have now would be okay. More important then latency was actually throughput and high availability. And this was demonstrated by Aeropsike well.
Financially it was a good decision. The current cluster is a lot less than the original one, and traffic + data have grown since. we currently maintain 2 clusters of 5 i3.4xlarge machines. That's a total of 10 machines and is a lot cheaper of what we had before.
The DB is performing great. It is flash based and 99.98% of the queries have <1ms latency. Each XDR end holds around 3.1B records, with a replication factor of 2.
midterm load is around 3 (very low) and we are doing around 190K reads p/s plus 37K write p/s at pick load.