https://tracker.ceph.com/https://tracker.ceph.com/favicon.ico2011-10-20T16:18:31ZCeph Ceph - Bug #1636: reweight-by-utilization does not choose good weightshttps://tracker.ceph.com/issues/1636?journal_id=67352011-10-20T16:18:31ZJosh Durgin
<ul><li><strong>Subject</strong> changed from <i>distributing accross devices with different capacity</i> to <i>reweight-by-utilization does not choose good weights</i></li><li><strong>Category</strong> set to <i>OSD</i></li><li><strong>Assignee</strong> set to <i>Josh Durgin</i></li><li><strong>Target version</strong> set to <i>v0.38</i></li></ul> Ceph - Bug #1636: reweight-by-utilization does not choose good weightshttps://tracker.ceph.com/issues/1636?journal_id=67362011-10-20T17:45:14ZJosh Durgin
<ul><li><strong>Status</strong> changed from <i>New</i> to <i>Resolved</i></li></ul><p>The existing reweight-by-utilization code didn't make sense - <a class="changeset" title="OSDMonitor: reweight towards average utilization The existing reweight-by-utilization calculatio..." href="https://tracker.ceph.com/projects/ceph/repository/revisions/f94a44e688883f2db0971435a5333a8b60c77dec">f94a44e688883f2db0971435a5333a8b60c77dec</a> fixes this.</p>
<p>When you have osds with different amounts of storage, you're better off setting the weights in the crushmap [[<a class="external" href="http://permalink.gmane.org/gmane.comp.file-systems.ceph.devel/2858">http://permalink.gmane.org/gmane.comp.file-systems.ceph.devel/2858</a> like this]]. This way you don't have to wait until some osds fill up. reweight-by-utilization is meant more for correcting for unfortunate random distribution.</p>