Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 15 Jun 2000 22:18:11 -0700 (PDT)
From:      Marc Slemko <marcs@znep.com>
To:        Nate Williams <nate@yogotech.com>
Cc:        Daniel O'Connor <doconnor@gsoft.com.au>, "Daniel C. Sobral" <dcs@newsguy.com>, cvs-all@FreeBSD.ORG, cvs-committers@FreeBSD.ORG, Alfred Perlstein <alfred@FreeBSD.ORG>
Subject:   Re: cvs commit: src/sys/kern uipc_socket.c uipc_socket2.c src/sy
Message-ID:  <Pine.GSO.4.21.0006152210490.14352-100000@redfish>
In-Reply-To: <200006160506.XAA28743@nomad.yogotech.com>

next in thread | previous in thread | raw e-mail | index | archive | help


On Thu, 15 Jun 2000, Nate Williams wrote:

> > >  HTTP belongs to userland. We should no more add support to HTTP than we
> > >  should add support to the dozens of different userland streaming
> > >  protocols in existance.
> > 
> > So we should just not bother squeezing more performance out of the system on
> > architectural grounds?
> 
> Correct.  If we want performance, how about removing the difference
> between kernel and userland.  Why have the overhead of the kernel
> vs. userland change, since it causes us to take a big performance hit.
> 
> As long as we're at it, we may as well embed all of the video drivers in
> the kernel, like NT did.  X should become an embedded part of the
> system, since that will also speed up performance again.
> 
> There are *LOTS* of performance improvments that can be done.  Since
> FreeBSD is used on the internet, let's embed telnet and ftp in the
> kernel as well.  We could certainly blow away all benchmarks that people
> see.

I think the real point here is that, if this sort of specialized
performance is required, then great, people can do it as a random LKM
or whatever that drops in using reasonably generic and appropriate
hooks.  Most of the time, that can be done.  It can be done for HTTP.
If someone needs video performance at a level that can only be
accomplished by moving their drivers into kernel space, I have no
problems with them doing that.

It is all a matter of adding it to the right places to make it fairly
drop in.  IBM added a kernel HTTP cache to NT using, if I recall
correctly, no code changes (or source) to NT, only pre-existing hooks.
Maybe I think that is useful.  Maybe I don't.  But there is every 
reason to support the flexibility it givers as long as it is done in
a way that lets those who want it drop it in cleanly.  Maybe all the
required hooks are there.  Maybe they can be cleanly added.  Either
way.



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.GSO.4.21.0006152210490.14352-100000>