Github user roshannaik commented on the issue:
https://github.com/apache/storm/pull/2203
Just wanted to unblock things here ... will post summarized numbers from my
runs by tomorrow.... here is a very brief summary.
I ran both TVL and ConstSpoutIdBoltNullBoltTopo (with and without ACKing)
in single worker mode.
**With ACKING**: The latency and throughput numbers are slightly better
for this patch.
**Without ACKING**: In case of both binaries, the worker dies & restarts
after a few every minutes of running if we let the topo run without any
throttling. (a known issue that will get fixed in STORM-2306 for Storm 2.0) The
numbers taken when the worker is running generally indicate that this patch is
most likely slower wrt the peak throughput. IMO, For Storm1.x this is not an
issue as at higher throughputs worker is going to keep dying. But for Storm
2.0, once we have STORM-2306, it will be possible to better measure the
throughput and a fix may be necessary.
In short, my runs indicate that things are looking good for this patch wrt
Storm 1.x. May need to revisit perf for Storm 2.0
As mentioned before, will post summarized numbers later.
---
