Linux performance enhancement

[Rafael Umann]

Hi Poul,

The problem with this patch is that it keeps too many sockets opened.
In a high load server (maybe with load balancers), it can be a potential
problem, and much worst in atacks!

So, we would like to know your (and anybody else) opinion about it.
What is better?

1. Performance with lots of sockets waiting 60 seconds to be cleaned.
2. Performance with the normal amount of sockets, with a thread
controling all this socket stuff.
3. No performance :(

Anyway, with this patch (option 1) the performance is ok, but it also
implement some problems. And we could not use varnish with option 3.

For us, the ideal solution would be implementing a new thread to
control all this stuff. What do you think about it?

Thanks,

Br,
-- 
Rafael Umann <rafael.umann at corp.terra.com.br>
Suporte Engenharia 1
Terra Networks Brasil S/A 
Tel: +55 (51) 3284-4344


On Mon, 02 Jun 2008 19:32:46 +0000
"Poul-Henning Kamp" <phk at phk.freebsd.dk> wrote:

> In message <20080602160205.633e479c at sup-dig.wrk.terra.com.br>, Rafael
> Umann wri tes:
> 
> >We've a ticket and would like to hear anything from you:
> >
> >http://varnish.projects.linpro.no/ticket/235
> 
> Hi Rafael,
> 
> I don't have linux machines in my lab to test this on, so I was hoping
> to get Dag-Erling to give it a spin.
> 
> He is on another project right now, so that is probably not happening.
> 
> So if you tell me the patch is guaranteed to work better than what's 
> in trunk, I'll commit it.
>