xfs
[Top] [All Lists]

Re: On RAID, inode size, stripe size (was: Playing around with NFS+XFS)

To: Federico Sevilla III <jijo@xxxxxxxxxxxxxxxxxxxx>
Subject: Re: On RAID, inode size, stripe size (was: Playing around with NFS+XFS)
From: Steve Lord <lord@xxxxxxx>
Date: Wed, 05 Sep 2001 10:36:24 -0500
Cc: Dan Yocum <yocum@xxxxxxxx>, Linux XFS Mailing List <linux-xfs@xxxxxxxxxxx>, "Philippine Linux Users' Group Mailing List" <plug@xxxxxxxxxxxxxxxxx>
In-reply-to: Message from Federico Sevilla III <jijo@xxxxxxxxxxxxxxxxxxxx> of "Wed, 05 Sep 2001 23:19:40 +0800." <Pine.LNX.4.33.0109052304430.20382-100000@xxxxxxxxxxxxxxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
> 
> Would you mind explaining to my young mind how using an inode size of 512
> bytes protects from data/inode corruption with hardware RAID5? Will this
> be significant for other setups (hardware RAID10, software RAID, no RAID
> at all)? Does this have any major disk space or performance impacts?
> 

This is not a raid5 thing, it is a filesystem size issue, once you get
above 1 Tbyte in filesystem size then xfs inode numbers (which are really
a disk address) can take more than 32 bits. Since lots of linux code,
including NFS, does not cope with this, we need to change things in xfs
so that a larger inode is used, this reduces the number of addressing
bits required down to below 32 bits again.

This is an interim measure, we have a design for keeping inodes down in the
low part of the filesystem (low Tbyte that is) to avoid the address space
overflow.

Steve



<Prev in Thread] Current Thread [Next in Thread>