Suspecting that our single jmeter client was the bottleneck, we experimented with running multiple clients, but on one machine.
And surprisingly throughput was easily doubled where we previously hit a wall.
What is the bottleneck here? I assumed JMeter could use one machine to the fullest. Is there some configuration I can tweak (tried increasing the heap size, didn't help)
Total CPU and memory utilization stayed the same (<25%). Is there something about networking on centOS 7 that restricts the throughput for a single process? Or did we misconfigure JMeter somehow? We are talking about 1k queries per second over JDBC from 200-300 threads. This should be doable by one JMeter instance, right?