svn commit: samba r22132 - in branches: SAMBA_3_0/source/modules SAMBA_3_0/source/smbd SAMBA_3_0_25/source/modules SAMBA_3_0_25/source/smbd

[James Peach]

On 08/04/2007, at 4:54 PM, Jeremy Allison wrote:

> On Sun, Apr 08, 2007 at 03:25:55PM -0700, James Peach wrote:
>> On 08/04/2007, at 12:41 PM, jra at samba.org wrote:
>>
>>> Author: jra
>>> Date: 2007-04-08 19:41:47 +0000 (Sun, 08 Apr 2007)
>>> New Revision: 22132
>>>
>>> WebSVN: http://websvn.samba.org/cgi-bin/viewcvs.cgi?
>>> view=rev&root=samba&rev=22132
>>>
>>> Log:
>>> I hate inefficiency. Don't call conv_str_size()
>>> on every pread/sendfile call, initialize these
>>> variables in an allocated struct at connection
>>> time and just refer to them directly.
>>
>> This code is so close to vfs_cacheprime that I'm tempted to suggest
>> they should be merged. Was there a particular reason you didn't do  
>> this?
>
> Yes - mine is much simpler :-). I was just doing
> the minimal thing that works for Vista.
>
> Doing cacheprime isn't the same as the readahead()
> system call on Linux. posix_fadvise is close enough
> that I could squeeze it in....

You could add this to vfs_cacheprime, and just have an option that  
determines whether it does a readahead (ie. I/O hint) or an actual  
read(2).

> This code gives a marked speed increase for Vista
> clients when using explorer against Linux servers.

Sure, but 2 modules that do *almost* the same thing is just confusing.

--
James Peach | jpeach at samba.org