Github user revans2 commented on the pull request:
https://github.com/apache/storm/pull/568#issuecomment-107976283
@eshioji,
I ran some speed of light [tests](https://github.com/yahoo/storm-perf-test)
on my macbook pro, that was running all of the processes, and had 4 slots
configured. I ran with 4 spouts, 8 bolts, acking enabled and I varied the
maxSpoutPending with 10, 1000, and 2000. I calculated the average number of
messages per millisecond.
maxSpoutPending | 0.11.0-SNAPSHOT | STORM-763 | diff
----------------|---------|--------------------------|-------
10 | 28.6491 | 26.9551 | -1.694 (-5.9%)
1000 | 45.2541 | 43.0215 | -2.2326 (-4.9%)
2000 | 42.7067 | 40.5542 | -2.1525 (-5.0%)
As you can see this change is having an impact of about -5% on the
performance in this one use case. I do like very much that the code is smaller
and STORM-763 seems significant enough that I am inclined to let this in as is,
but I really would like to have you take a crack at possibly doing some
optimizations to the code. I'm not sure but my guess is that it is the 64k
high water mark, which is switching the channel to no longer be writable, or
the fact that we do not batch in-between calls to send.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
