Hi Cassandra community,
we conducted performance benchmarking experiments on Cassandra 0.7 (on
EC2 instances) using YCSB++ and stumbled upon some unexpected results:
1. Changing consistency level configurations from Write.ALL + Read.ONE
to Write.ALL + Read.ALL increases write latency (expected) and
decrease read latency (unexpected).
2. Changing from a single-region Cassandra cluster to a multi-region
Cassandra cluster on EC2 significantly increases write latency (approx
2x, again expected) but slightly decreases read latency (approx -10%,
again unexpected).
The effects could be due to unexplained variance in our model (we
might have forgotten to include important independent variables),
however we got quite good "R-squared adjusted" and are therefore
somewhat confident that the results reflect some mechanism in
Cassandra. Do you have an explanation for the effect? It seems
counter-intuitive at first sight.
Thanks,
Markus
|