Github user HeartSaVioR commented on the issue:
https://github.com/apache/storm/pull/2241
@avermeer
As @roshannaik stated, we addressed many parts of performance bottleneck on
Storm 1.0.0. The paper about Heron and blog post on SuperChief referred old
version of Storm so it doesn't even hold for 1.0.0.
(Yes I understand they can't pick 1.0.0 to compare since 1.1.0 is released
at Apr. 2016.)
I think we learnt a lot and adopt some parts of them like back pressure
from Heron's paper (though we implemented in different way), but personally
also wonder of their points of performance gaining, especially 1 computational
thread on 1 worker. Message should be serialized even across the worker in a
node, and the cost of serialization/deserialization is not that cheap.
I think there's no general way of benchmark in Streaming, so claiming that
A is N times faster than B could hide too much detail. Most of frameworks has
their own characteristics hence excels on specific use case or performance
test. Moreover, developer from A doesn't know about B deeply so most of time
fails to set variables optimized specifically for the test.
That's why I prefer comparing with our own history much instead of
comparing it with others.
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
