[
https://issues.apache.org/jira/browse/STORM-815?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14539876#comment-14539876
]
Robert Joseph Evans commented on STORM-815:
-------------------------------------------
[~rsprabery],
I love the idea of having storm be SLA aware, and would love to see any
research and or code that you have already done on the topic. I think there is
a lot of fertile ground within storm to explore topology elasticity, resource
awareness and resource utilization for the cluster. Allowing topology owners
to express their desired SLA opens up a lot of possibilities for automatically
tuning the topology to meet both desired latency, throughput, and resource
utilization and the relative importance of each. We have been working with a
few other researchers looking at resource aware scheduling and I think the two
would complement each other very well.
> Auto tuning topologies based on user submitted SLA
> --------------------------------------------------
>
> Key: STORM-815
> URL: https://issues.apache.org/jira/browse/STORM-815
> Project: Apache Storm
> Issue Type: New Feature
> Reporter: Read Sprabery
> Priority: Blocker
>
> Hi all – not sure if this is the place for this or not, so please feel free
> to move this ticket around.
> I have been working on making Storm SLA aware and wanted to create a ticket
> focused on auto-tuning of submitted topologies. Having talked with [~revans2]
> , I think this feature will go a long way in improving the usability of Storm.
> This will be particularly useful as more abstractions are developed on top of
> Storm and developers move further away from the internals of topologies with
> systems like Summingbird (https://github.com/twitter/summingbird). Storm
> already has the majority of the metrics required to made decisions regarding
> the stability of a given stream in Zookeeper; these can be used to decide if
> a given SLA is being met.
> The system I'm working on functions externally to Storm, getting measurements
> from the Thrift interface and then performing parallelism level tuning
> through the “rebalance” command. If the community is interested, I can upload
> a document in the upcoming weeks highlighting the challenges and benefits of
> an auto-tuning system in Storm and that can serve as a first start for
> developing an auto-tuning scheduler.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
