[
https://issues.apache.org/jira/browse/STORM-297?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14017695#comment-14017695
]
ASF GitHub Bot commented on STORM-297:
--------------------------------------
Github user Gvain commented on the pull request:
https://github.com/apache/incubator-storm/pull/103#issuecomment-45091025
@clockfly
Thanks for your explanation. The explanation for part 2 seems not so
convincing as the CPU and network are not saturated at both env when scaling
worker# from 4 to 8. Oops, I don't wanna go any further on this.
Anyway, your work is really great.
Another thing. Since you use netty to send message in a async way, how can
you ensure that all messages emitted from a fast component are received by a
slow component without any loss especially under heavy throughput without any
ackers enabled ? To my knowledge, the netty channel may not be writable, and
pendings count will increase.
> Storm Performance cannot be scaled up by adding more CPU cores
> --------------------------------------------------------------
>
> Key: STORM-297
> URL: https://issues.apache.org/jira/browse/STORM-297
> Project: Apache Storm (Incubating)
> Issue Type: Bug
> Reporter: Sean Zhong
> Labels: Performance, netty
> Fix For: 0.9.2-incubating
>
> Attachments: Storm_performance_fix.pdf,
> storm_Netty_receiver_diagram.png, storm_conf.txt,
> storm_performance_fix.patch, worker_throughput_without_storm-297.png
>
>
> We cannot scale up the performance by adding more CPU cores and increasing
> parallelism.
> For a 2 layer topology Spout ---shuffle grouping--> bolt, when message size
> is small (around 100 bytes), we can find in the below picture that neither
> the CPU nor the network is saturated. When message size is 100 bytes, only
> 40% of CPU is used, only 18% of network is used, although we have a high
> parallelism (overall we have 144 executors)
--
This message was sent by Atlassian JIRA
(v6.2#6252)
