[ 
https://issues.apache.org/jira/browse/HBASE-14604?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Guanghao Zhang updated HBASE-14604:
-----------------------------------
    Status: Open  (was: Patch Available)

> Improve MoveCostFunction in StochasticLoadBalancer
> --------------------------------------------------
>
>                 Key: HBASE-14604
>                 URL: https://issues.apache.org/jira/browse/HBASE-14604
>             Project: HBase
>          Issue Type: Bug
>          Components: Balancer
>    Affects Versions: 0.98.15
>            Reporter: Guanghao Zhang
>            Assignee: Guanghao Zhang
>         Attachments: HBASE-14604-0.98.patch, HBASE-14604-0.98_v1.patch, 
> HBASE-14604-trunk.patch, HBASE-14604-trunk_v1.patch, HBASE-14604_98.diff, 
> HBASE-14604_98_with_ut.diff
>
>
> The code in MoveCoseFunction:
> {code}
> return scale(0, cluster.numRegions + META_MOVE_COST_MULT, moveCost);
> {code}
> It uses cluster.numRegions + META_MOVE_COST_MULT as the max value when scale 
> moveCost to [0,1]. But this should use maxMoves as the max value when cluster 
> have a lot of regions.
> Assume a cluster have 10000 regions, maxMoves is 2500, it only scale moveCost 
> to [0, 0.25].
> Improve moveCost by use Math.min(cluster.numRegions, maxMoves) as the max 
> cost, so it can scale moveCost to [0,1].
> {code}
> return scale(0, Math.min(cluster.numRegions, maxMoves) + META_MOVE_COST_MULT, 
> moveCost);
> {code}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to