Github user roshannaik commented on the issue:
https://github.com/apache/storm/pull/2203
@ptgoetz, I suggest measuring the perf impact with ACKers=0 as well (with
workers=1). Having ACKing enabled or interworker-communication going on hides
many perf issues.
In addition to TVL, it is important to measure perf with other topos like
[ConstSpoutNullBoltTopo](https://github.com/apache/storm/blob/master/examples/storm-perf/src/main/java/org/apache/storm/perf/ConstSpoutNullBoltTopo.java)
and
[ConstSpoutIdBoltNullBoltTopo.java](https://github.com/apache/storm/blob/master/examples/storm-perf/src/main/java/org/apache/storm/perf/ConstSpoutIdBoltNullBoltTopo.java).
The ConstSpout* topos can expose perf issues that TVL might not... and vice
versa.
---
