Github user roshannaik commented on the issue:
https://github.com/apache/storm/pull/2270
Ok thats great since I noticed that the -c mode was not working for the
grouping option due a bug in the topo creation code .
BTW... I noticed that I incorrectly said something important earlier that I
want to correct... actually the ACKer is slower than the inter-worker (and not
the other way around as I accidentally said).
Your observations about LoadAware are similar to mine ... which i
discovered when doing STORM-2306.
With STORM-2306, both ACKer and inter-worker throughputs are a good deal
higher (but nowhere close to where they really should be).. So I felt it would
be interesting to run this analysis on it to observe impact. Since this patch
doesn't apply easily on 2306, it took me a bunch of time to manually resolve
it.
Anyway here are the results:
**No ACK mode**
LocalOrShuffle: ~**7.12 mill/sec**
LASG : ~ **6.5 mill /sec**
**ACK mode**
LocalOrShuffle: ~ **1.23 mill/sec** . Latency: avg: **0.013 ms** max:
**0.015 ms**
LASG : ~ **1.26 mill/sec** Latency: avg: **0.017 ms**
max: **0.035 ms**
8% slower in noACK mode. Avg latency is close, max latency is 2x+ higher
in ACK mode. But should be ok.
Overall these numbers look acceptable to me.
Once (hopefully soon) we fix the two bottlenecks (ACKer & interworker),
there is a possibility that we may have to revisit and optimize this.
Have you had a chance to test this on a real multi-rack cluster and make
any observations ?
---
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.
---
