From: "Roger Heflin" <rheflin@atipa.com>
Subject: RE: NFS tuning - high performance throughput.
Date: Wed, 15 Jun 2005 15:31:30 -0500
Message-ID: <EXCHG2003tosJ3Ih7jt000005f9@EXCHG2003.microtech-ks.com>
References: <1118866263.22484.21.camel@seki.nac.uci.edu>
Mime-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Cc: "'M. Todd Smith'" <todd@sohovfx.com>, <nfs@lists.sourceforge.net>
To: "'Dan Stromberg'" <strombrg@dcs.nac.uci.edu>
In-Reply-To: <1118866263.22484.21.camel@seki.nac.uci.edu>
Sender: nfs-admin@lists.sourceforge.net
Errors-To: nfs-admin@lists.sourceforge.net

 
If you do full sync things are horribly slow.

If you are doing it in C code, a single sync kernel call
works nicely before the close, to flush everything.

                       Roger

> -----Original Message-----
> From: Dan Stromberg [mailto:strombrg@dcs.nac.uci.edu] 
> Sent: Wednesday, June 15, 2005 3:11 PM
> To: Roger Heflin
> Cc: 'M. Todd Smith'; nfs@lists.sourceforge.net; 
> strombrg@dcs.nac.uci.edu
> Subject: RE: [NFS] NFS tuning - high performance throughput.
> 
> On Wed, 2005-06-15 at 14:52 -0500, Roger Heflin wrote:
> > The mount/remount (unless you time how long it takes to 
> unmount), will 
> > potentially still skew the write test, and the writes will still be 
> > happening for quite a while after the dd completes.
> 
> Or unless you sync twice?  Or write synchronously.
> 
> :)
> 
> >                         Roger
> > 
> > > -----Original Message-----
> > > From: Dan Stromberg [mailto:strombrg@dcs.nac.uci.edu]
> > > Sent: Wednesday, June 15, 2005 2:14 PM
> > > To: Roger Heflin
> > > Cc: 'M. Todd Smith'; nfs@lists.sourceforge.net; 
> > > strombrg@dcs.nac.uci.edu
> > > Subject: RE: [NFS] NFS tuning - high performance throughput.
> > > 
> > > 
> > > A bigger dd test should work - if you use about 3x your physical 
> > > memory, you'll probably be fine doing it that way.
> > > 
> > > Another way that may work better in -some- cases, is to 
> umount the 
> > > filesystem and re-mount it - except for warm restartable 
> > > filesystems.
> > > 
> > > On Wed, 2005-06-15 at 10:28 -0500, Roger Heflin wrote:
> > > > Make the dd test bigger, with 2x the memory size, 50% 
> will be from 
> > > > cache and almost instanteous, with 4x the memory size, 
> 25% will be 
> > > > from cache, bigger runs will get closer to actual 
> reality, on the 
> > > > tests I do I run things 8x, but if you run 2x, 4x, 8x 
> memory you 
> > > > should get a decent graph to give you an idea of what the
> > > actual value
> > > > is.
> > > > 
> > > >                Roger
> > > > 		   Atipa Technologies. 
> > > > 
> > > > > -----Original Message-----
> > > > > From: nfs-admin@lists.sourceforge.net 
> > > > > [mailto:nfs-admin@lists.sourceforge.net] On Behalf Of M.
> > > Todd Smith
> > > > > Sent: Wednesday, June 15, 2005 9:47 AM
> > > > > To: nfs@lists.sourceforge.net
> > > > > Subject: Re: [NFS] NFS tuning - high performance throughput.
> > > > > 
> > > > > Roger Heflin wrote:
> > > > > 
> > > > > >Are you using the same dd test on the local machine test?  
> > > > > If so cache
> > > > > >will be a major factor.
> > > > > >
> > > > > >Also, raid5 stripe size, bigger is almost always better,
> > > I would do
> > > > > >some testing with different strip sizes and see how it
> > > affects the
> > > > > >speed, I have never seen less than 32k be faster than 32k.
> > > > > >
> > > > > >Are you using md for the raid5 setup or something else
> > > that has not
> > > > > >been mentioned?
> > > > > >
> > > > > >                     Roger
> > > > > >  
> > > > > >
> > > > > Roger,
> > > > > 
> > > > > I was told by a consultant we hired to help us with this that 
> > > > > 5Gb test files should be large enough to blow out the caches.
> > > > >  Looking back I realize that must have been when we only
> > > had 2Gb RAM
> > > > > in the machine ..
> > > > > is there a better test I can use to check local rw
> > > performance, as
> > > > > well is there a formula or recommended way to 
> calculate how big 
> > > > > a file you should rw to properly check local rw?
> > > > > 
> > > > > I'll have to ask our soft eng who worked with the
> > > consultant as to
> > > > > why that particular stripe size was used.  I believe we are 
> > > > > using md, we attempted to use LVM but ran into some problems
> > > and had some
> > > > > file inconsistencies that were unacceptable so had to 
> back out 
> > > > > of it.
> > > > > 
> > > > > Cheers
> > > > > Todd
> > > > > 
> > > > > --
> > > > > Systems Administrator
> > > > > ----------------------------------
> > > > > Soho VFX - Visual Effects Studio
> > > > > 99 Atlantic Avenue, Suite 303
> > > > > Toronto, Ontario, M6K 3J8
> > > > > (416) 516-7863
> > > > > http://www.sohovfx.com
> > > > > ----------------------------------
> > > > > 
> > > > > 
> > > > >  
> > > > > 
> > > > > 
> > > > > 
> > > > > -------------------------------------------------------
> > > > > SF.Net email is sponsored by: Discover Easy Linux Migration 
> > > > > Strategies from IBM. Find simple to follow Roadmaps,
> > > straightforward
> > > > > articles, informative Webcasts and more! Get everything
> > > you need to
> > > > > get up to speed, fast.
> > > > > http://ads.osdn.com/?ad_id=7477&alloc_id=16492&op=click
> > > > > _______________________________________________
> > > > > NFS maillist  -  NFS@lists.sourceforge.net 
> > > > > https://lists.sourceforge.net/lists/listinfo/nfs
> > > > > 
> > > > 
> > > > 
> > > > 
> > > > -------------------------------------------------------
> > > > SF.Net email is sponsored by: Discover Easy Linux Migration
> > > Strategies
> > > > from IBM. Find simple to follow Roadmaps, straightforward 
> > > > articles, informative Webcasts and more! Get everything 
> you need 
> > > > to get up to speed, fast. 
> > > > http://ads.osdn.com/?ad_id=7477&alloc_id=16492&op=click
> > > > _______________________________________________
> > > > NFS maillist  -  NFS@lists.sourceforge.net 
> > > > https://lists.sourceforge.net/lists/listinfo/nfs
> > > > 
> > > 
> > 
> 


-------------------------------------------------------
SF.Net email is sponsored by: Discover Easy Linux Migration Strategies
from IBM. Find simple to follow Roadmaps, straightforward articles,
informative Webcasts and more! Get everything you need to get up to
speed, fast. http://ads.osdn.com/?ad_id=7477&alloc_id=16492&op=click
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs