2.5 readpages doubles cifs vfs large file copy performance vs. 2.4

[David Collier-Brown -- Customer Engineering]

   Absolutely, it's just that the usage pattern that is
optimal on NT is suboptimal on RAID. It's also **really**
suboptimal on ufs, but that's my cross to bear (;-))

   I think of Samba as a big impedance-match device: one
of the mismatches happens to be the amount of readahead
when used with certain filesystems or disk arrays.

   I would prefer to see it fixed at the OS level, as the OS
should know the right parameters from the disk devices,
with a second choice of using setvbuf to hint when the
OS doesn't actually know the impedance, or knows a usable
but inadvisable value.
   Alas, setvbuf only works when we're using stdio. Since we
use read(2), we don't get to use it...

--dave

Jeremy Allison wrote:
> On Thu, Jul 03, 2003 at 09:42:58AM -0400, David Collier-Brown -- Customer Engineering wrote:
> 
>>   Observation with truss (the Solaris strace) showed Samba doing
>>a read of 1 KB, then 16 KB and finally of  64KB, which was
>>**pessimal** for our ufs (which is really an improved BSD ffs).
> 
> 
> Yeah, but it's the size the client is asking for. That isn't
> something we decide.
> 
> Jeremy.
> 


-- 
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 canada.sun.com