I try to test the throughput using both the High Level Consumer and Simple Consumer example from the document. But I get much lower throughput of simple consumer than the high level consumer. I run the test in the cluster and I'm sure I distribute the leader broker and consumer reads on different machine. So Is there any suggestion for the high throughput?
Another question is if produce rate of producer is much higher than consumer, what would happen? All the data would pile up in the broker?