Github user zhuoliu commented on the pull request:
https://github.com/apache/storm/pull/700#issuecomment-139678312
2. Test with a congested topology.
WordCountTopology2 (3 times more workers and executors than original
WordCount; in every minute, for the wordcount bolt: in the first 30 second,
there is a 50 ms sleep before each tuple emits"if (time/30000 %2 == 0)
Thread.sleep(50);").
Interestingly, without ABP, this 12 worker (87 executors) topology can not
run successfully in Storm since workers frequently crash because overflowing of
tuples in the bolt executors.
And the ZK receive workload is 20.6 to 21 pack/second.
On the contrast, with ABP enabled, this topology runs pretty well.
And we see about 22.5 pack/sec receive workloads on Zookeeper nodes.
This test shows the great advantage of Backpressure when dealing with
topologies that may have congested or slow components. Since (1) ABP makes sure
this topology can run successfully; (2) ABP causes small overheads to Zookeeper.
@revans2 @d2r @knusbaum
---
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.
---
