O_SYNC behavior?

O_SYNC behavior?

[Ray Van Dolson]

End-user question:

NFS client accessing an NFS export on an XFS based filesystem via
"sync" (O_SYNC) mode.

Will write confirmation be sent back from the NFS server when the data
is written to the xfs journal, or not until it is actually written to
disk (assuming in this case the journal is on a separate SSD device or
something similar)?

Thanks,
Ray

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: O_SYNC behavior?

[Nathan Scott]

----- "Ray Van Dolson" <rvandolson@esri.com> wrote:

> End-user question:
> 
> NFS client accessing an NFS export on an XFS based filesystem via
> "sync" (O_SYNC) mode.
> 
> Will write confirmation be sent back from the NFS server when the
> data is written to the xfs journal,

Data is never written to the journal.

cheers.

-- 
Nathan

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: O_SYNC behavior?

[Matthias Schniedermeyer]

On 27.04.2010 09:25, Nathan Scott wrote:
> 
> ----- "Ray Van Dolson" <rvandolson@esri.com> wrote:
> 
> > End-user question:
> > 
> > NFS client accessing an NFS export on an XFS based filesystem via
> > "sync" (O_SYNC) mode.
> > 
> > Will write confirmation be sent back from the NFS server when the
> > data is written to the xfs journal,
> 
> Data is never written to the journal.

I don't think that answers the intented question.

I guess the correct answer is:
It depends on the server-side configuration.
An "async" export gets the "done"-answer immediatly.
An "sync" export (only when also mounted "sync" AFAIU) waits at least 
until the write command is send down the stack, maybe even until the 
data actually hit the platters or silicon in the SSD-case.

man 5 exports:
- snip -
sync:	Reply to requests only after the changes have been committed to stable storage
- snip -

But i guess the performance will be horrible when both side are "sync".




Bis denn

-- 
Real Programmers consider "what you see is what you get" to be just as 
bad a concept in Text Editors as it is in women. No, the Real Programmer
wants a "you asked for it, you got it" text editor -- complicated, 
cryptic, powerful, unforgiving, dangerous.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: O_SYNC behavior?

[Christoph Hellwig]

On Tue, Apr 27, 2010 at 11:32:50AM +0200, Matthias Schniedermeyer wrote:
> I don't think that answers the intented question.
> 
> I guess the correct answer is:
> It depends on the server-side configuration.
> An "async" export gets the "done"-answer immediatly.
> An "sync" export (only when also mounted "sync" AFAIU) waits at least 
> until the write command is send down the stack, maybe even until the 
> data actually hit the platters or silicon in the SSD-case.
> 
> man 5 exports:
> - snip -
> sync:	Reply to requests only after the changes have been committed to stable storage
> - snip -
> 
> But i guess the performance will be horrible when both side are "sync".

The client side (that is mount-side) sync means implying O_SYNC for all
I/O - no need for that.  And sync exports are the default for every sane
server.

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: O_SYNC behavior?

[Ray Van Dolson]

On Tue, Apr 27, 2010 at 07:12:05AM -0700, Christoph Hellwig wrote:
> On Tue, Apr 27, 2010 at 11:32:50AM +0200, Matthias Schniedermeyer wrote:
> > I don't think that answers the intented question.
> > 
> > I guess the correct answer is:
> > It depends on the server-side configuration.
> > An "async" export gets the "done"-answer immediatly.
> > An "sync" export (only when also mounted "sync" AFAIU) waits at least 
> > until the write command is send down the stack, maybe even until the 
> > data actually hit the platters or silicon in the SSD-case.
> > 
> > man 5 exports:
> > - snip -
> > sync:	Reply to requests only after the changes have been committed to stable storage
> > - snip -
> > 
> > But i guess the performance will be horrible when both side are "sync".
> 
> The client side (that is mount-side) sync means implying O_SYNC for all
> I/O - no need for that.  And sync exports are the default for every sane
> server.
> 

Thanks guys.  I was asking this from the perspective of a ZFS type
setup with its intent log (ZIL) on SSD's or some battery backed NVRAM.

In this setup, with sync mode NFS, acknowledgement is sent to the
client as soon as the data is written to the ZIL -- before it actually
hits the underlying disks.

Just was trying to determine a way to mirror this setup with XFS.

Thanks,
Ray

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs