[ 
https://issues.apache.org/jira/browse/ACCUMULO-3083?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Christopher Tubbs resolved ACCUMULO-3083.
-----------------------------------------
    Resolution: Abandoned

Closing this stale issue. If this is still a problem, please create a new issue 
or PR at https://github.com/apache/accumulo

> capability balancing
> --------------------
>
>                 Key: ACCUMULO-3083
>                 URL: https://issues.apache.org/jira/browse/ACCUMULO-3083
>             Project: Accumulo
>          Issue Type: New Feature
>          Components: master
>            Reporter: Eric C. Newton
>            Priority: Major
>
> I was thinking about balancing, in much the way that new container-based 
> (yarn, mesos, etc) resource managers work.  The existing {{Balancers}} try to 
> keep the number of tablets even over presumably homogeneous nodes.
> But as clusters age and are improved and expanded in piecemeal, the hardware 
> capabilities will drift apart.  Some nodes may be capable of handling more 
> load than others.  We may even want some nodes to have a trivial number of 
> tablets so the data they serve is always cached.
> Tablet servers could advertise an attribute (or several).  Really, it would 
> just be a name, like {{fatnode}} or {{metaonly}}.  These would then be used 
> by the {{Balancer}} to favor or ignore tservers for balancing.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to