[PATCH 2/3] os: Don't limit MaxClients to MAXCLIENTS

[Adam Jackson]

On Tue, 2009-12-15 at 08:55 -0800, Keith Packard wrote:
> On Tue, 15 Dec 2009 09:32:58 -0500, Adam Jackson <ajax at nwnk.net> wrote:
> 
> > Well, not devices, at least not for silken-enabled servers.  But yes.
> > And right now, it does; MaxClientFd ends up being getdtablesize()-1,
> > which is all the file descriptors you can pass to select.
> 
> I was just thinking that 'MaxClientFd' was a confusing name as it's
> possible for some non-client fd to be present in the array.

It is.  Corrected patch to follow momentarily.

> In any case, I'd bet that x11perf -prop or x11perf -pointer would be
> better able to detect any difference here; you really want to hammer the
> select syscall, and -noop won't end up calling that very often.

Quick experiment shows <1% variation for both -prop and -pointer
performance on Xvfb (without this patch series) with MaxClients at any
of 16, 256, and 1024.  That's close enough to experimental noise for me.

Of course, to get that number, I had to patch x11perf to not truncate
results to 3 significant figures (which necessarily limits any variance
measurement to 1% granularity).  I'll probably push the patch for that,
although off by default and enabled with -false-precision so the
phoronix types don't get the wrong idea.

- ajax
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: This is a digitally signed message part
Url : http://lists.x.org/archives/xorg-devel/attachments/20091215/7be3e904/attachment.pgp