Comments on: VMFS Metadata size? http://www.yellow-bricks.com/2009/11/11/vmfs-metadata-size/ Building blocks for virtualization... Fri, 19 Mar 2010 10:49:18 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Duncan Epping http://www.yellow-bricks.com/2009/11/11/vmfs-metadata-size/comment-page-1/#comment-5041 Duncan Epping Thu, 12 Nov 2009 08:31:42 +0000 http://www.yellow-bricks.com/?p=4509#comment-5041 Thanks Doug, looked into it and edited my article. Thanks Doug, looked into it and edited my article.

]]> By: dougied85 http://www.yellow-bricks.com/2009/11/11/vmfs-metadata-size/comment-page-1/#comment-5035 dougied85 Wed, 11 Nov 2009 21:56:56 +0000 http://www.yellow-bricks.com/?p=4509#comment-5035 I work for a company that has fallen foul of this calculation - in fact we are almost certainly the reason this has now been removed from the Design & Deployment Guide. This calculation is incorrect. We ended up in a situation where LUNs became inaccessible because they became 100% full as a result of a bug in the VMFS3 module that filled up the space we had reserved for metadata. This is also probably where the Advanced Option for OpenWithoutJournal came from, as a custom fix we were provided by Engineering gave us that option on our 3.01 environment at the time. I have been back and forth with VMware quite a lot over this issue, and the best I have managed to get out of them to date is that you should allow 1200MB per LUN (no matter what size of LUN) for metadata. They won't publish this officially though, as they feel LUN sizing is too dependent on other factors for them to be able to set official minimums that covers all. I work for a company that has fallen foul of this calculation – in fact we are almost certainly the reason this has now been removed from the Design & Deployment Guide.

This calculation is incorrect.

We ended up in a situation where LUNs became inaccessible because they became 100% full as a result of a bug in the VMFS3 module that filled up the space we had reserved for metadata. This is also probably where the Advanced Option for OpenWithoutJournal came from, as a custom fix we were provided by Engineering gave us that option on our 3.01 environment at the time.

I have been back and forth with VMware quite a lot over this issue, and the best I have managed to get out of them to date is that you should allow 1200MB per LUN (no matter what size of LUN) for metadata. They won’t publish this officially though, as they feel LUN sizing is too dependent on other factors for them to be able to set official minimums that covers all.

]]>