First ext3 data (was Re: 2.5 readpages doubles cifs vfs large file copy performance vs. 2.4)

David Collier-Brown -- Customer Engineering David.Collier-Brown at Sun.COM
Wed Jul 9 11:45:03 GMT 2003

And the second chunk of data is now in, for
single scsi disks wit hthe ext3 filesystem, and
varying amounts of read cache buffer.

The scsi transfer times are pretty flat, once you
get away from the default (shown as 0 in the spreadsheet
graphs). For 4 and 8 MB files, the times improved, for
16 MB files the times were the same, and for
32 MB files the times degraded when the buffer sizes
were moved away from the default. I'm going to
re-measure the 32 MB case, but

As always, throughput benefited, and as with ext3 and
IDE, it was mostly at the high end of the buffer
settings, even with a single disk, as opposed to
an array.

My right-off-the-top conclusions? Linux buffering
is surprisingly good on single disks, and the
"sweet spot" for user-side buffering is quite wide.


David Collier-Brown -- Customer Engineering wrote:
> David Collier-Brown -- Customer Engineering wrote:
>  >  The data I have from qfs and ufs on Solaris says bigger is better
>  > until you get too big, but it's fairly insensitive once you get
>  > past too small.  The usual u-shaped curve, with a very wide
>  > "sweet spot".
>  >
>  >  If the sweet spot on Linux is large, it's going to be easy. If
>  > it's hard to hit, then I have to start looking carefully at the
>  > criteria.
> I just did the initial test of increasing the filesystem
> read size on ext3 on an IDE disk, and there is indeed
> a moderate payoff in throughput for the case where
> the read size is  significant chunk of the file size,
> as you'd expect.
> What is interesting, however, is the shape of the
> absolute time curve. It has a very wide "sweet spot",
> starts to fall off at high buffer sizes, but that's
> where the small payoff really has value (in throughput),
> so it's credible to say that Linux EXT3 on a single
> IDE drive has a wide "sweet spot".
> Gifs attached: the different lines are for different
> file sizes, the other data is labeled. (hey, it's
> a spreadsheet graph, not an analysis (;-))
> --dave
> ------------------------------------------------------------------------
> ------------------------------------------------------------------------

David Collier-Brown,           | Always do right. This will gratify
Sun Microsystems DCMO          | some people and astonish the rest.
Toronto, Ontario               |
(905) 415-2849 or x52849       | davecb at

-------------- next part --------------
A non-text attachment was scrubbed...
Name: ext3scsi_time.gif
Type: image/gif
Size: 6102 bytes
Desc: not available
Url :
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ext3scsi_tput.gif
Type: image/gif
Size: 4886 bytes
Desc: not available
Url :

More information about the samba-technical mailing list