[
https://issues.apache.org/jira/browse/KAFKA-4816?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15935601#comment-15935601
]
Apurva Mehta commented on KAFKA-4816:
-------------------------------------
I compiled the perf numbers for this patch, results available here:
https://docs.google.com/spreadsheets/d/1dHY6M7qCiX-NFvsgvaE0YoVdNq26uA8608XIh_DUpI4/edit?usp=sharing
There is no observable regression on produce or consume at varying message
sizes, with different workloads, on different machines.
We do see significant on disk savings at smaller message sizes.
Given the nature of the tests, I think the margin of error in throughput
results is 10%. Results which show a difference of more than 10% should be
treated with a grain of salt: they generally don't reproduce across runs and
across machines.
We see consistent producer side performance improvements at small sizes,
correlated with lower produce request size.
We see general consume side improvements, but there is no pattern and there is
significant variation across runs. So we can't make any claims on improving the
consume performance with this message format. But we can claim safely that
there is no regression since all runs were generally faster with the new format.
> Message format changes for idempotent/transactional producer
> ------------------------------------------------------------
>
> Key: KAFKA-4816
> URL: https://issues.apache.org/jira/browse/KAFKA-4816
> Project: Kafka
> Issue Type: Sub-task
> Components: clients, core, producer
> Reporter: Jason Gustafson
> Assignee: Jason Gustafson
> Fix For: 0.11.0.0
>
>
> This task is for the implementation of the message format changes documented
> here:
> https://cwiki.apache.org/confluence/display/KAFKA/KIP-98+-+Exactly+Once+Delivery+and+Transactional+Messaging#KIP-98-ExactlyOnceDeliveryandTransactionalMessaging-MessageFormat.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
