xfs
[Top] [All Lists]

Re: stripe unit/width, inode64, misc mkfs.xfs params

To: Peter Merelis <merelis@xxxxxxxxxxxx>
Subject: Re: stripe unit/width, inode64, misc mkfs.xfs params
From: Eric Sandeen <sandeen@xxxxxxxxxxx>
Date: Fri, 30 Apr 2010 12:33:13 -0500
Cc: "xfs@xxxxxxxxxxx" <xfs@xxxxxxxxxxx>
In-reply-to: <133AD8A0-119B-4A41-9AF6-46F43111A941@xxxxxxxxxxxx>
References: <133AD8A0-119B-4A41-9AF6-46F43111A941@xxxxxxxxxxxx>
User-agent: Thunderbird 2.0.0.24 (Macintosh/20100228)
Peter Merelis wrote:
> Hi,
>  
> Apologies in advance as I imagine this issue is raised frequently, but can 
> someone familiar with these sorts of things sanity check my stripe unit/width 
> settings since the underlying storage controller is not exposing the data to 
> mkfs.xfs apparently:
>  
> 12 disk raid6
> 256k stripe
>  
> su=256k
> sw=10

sounds right

> …which I believe is equivalent to:
>  
> sunit=512
> swidth=5120
>  
> Is that correct?

probably, but FWIW su/sw might be easier since it deals in byte / multiplier 
units.


> Also, does the inode64 option make sense for a filesystem with only a few 
> files, each around ~100G?

for a large filesystem, yes, if your userspace can cope.

> And finally, are 'allocsize=1GB,nobarrier,noatime,nodiratime' recommended 
> mount options?

no mount options are recommended without more information about the load ;)

nobarrier is only recommended if the storage has a battery-backed cache (or no 
cache)

allocsize=1GB is probably ok if you really are only writing large files and have
fragmentation problems

noatime,nodiratime probably don't matter much at all.

-Eric

> Thanks,
> Peter
>  
> 
> _______________________________________________
> xfs mailing list
> xfs@xxxxxxxxxxx
> http://oss.sgi.com/mailman/listinfo/xfs
> 

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