xfs
[Top] [All Lists]

Re: xfs didn't provide redundancy for citical data structure?

To: hank peng <pengxihan@xxxxxxxxx>
Subject: Re: xfs didn't provide redundancy for citical data structure?
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Mon, 17 May 2010 09:32:43 +1000
Cc: xfs-oss <xfs@xxxxxxxxxxx>
In-reply-to: <AANLkTimP06rpMX4BRaRIggMctQ2qR7Cul9yqcJR6Xj3O@xxxxxxxxxxxxxx>
References: <AANLkTimP06rpMX4BRaRIggMctQ2qR7Cul9yqcJR6Xj3O@xxxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-06-14)
On Sat, May 15, 2010 at 11:57:51PM +0800, hank peng wrote:
> I read this paper: http://pages.cs.wisc.edu/~vshree/xfs.pdf, which
> says xfs didn't provide redundancy for citical data structure, such as
> mirror, parity. I wonder if it is so?
> If it is, is there plan to implement that for XFS developers?

XFS assumes redundancy and protection against bit errors is done in
the block layeri (i.e. RAID of some kind). Changing that assumption
takes a lot of work and involves modifying the disk format, so can't
be done overnight. That being said, work is in progress to make XFS
more robust - see this page for ideas on how we are approaching the
problem:

http://xfs.org/index.php/Reliable_Detection_and_Repair_of_Metadata_Corruption

I'll also point out that the above paper makes some fundamental
mistakes (e.g. XFS does not use "data=ordered" journalling as they
conclude it does from a limited observation). Hence the rest of
their results are somewhat questionable, too, as we can't examine
them closely enough to confirm or deny them....

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

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