2009-07-18 01:24:12

by Luis R. Rodriguez

[permalink] [raw]
Subject: khttpd fate

I was reviewing khttpd [1] history [2], and the last I see is it was
merged for the 2.4 kernel with enthusiasm. I don't see any information
about this for 2.6 though, nor can I find any other notes about why
this was removed. Just curious if someone recalls why it was removed.

Also, I really hate how trolly this questions sounds but here it goes anyway:

Such userspace-kernel hacks shouldn't be necessary anymore based on
'performance/latency' arguments right? I take it khttpd wasn't serious
but more of a hack for fun and now we should be able to laugh about
it?

[1] http://www.fenrus.demon.nl/
[2] http://lwn.net/2001/0118/kernel.php3

Luis


2009-07-22 00:20:43

by Kyle McMartin

[permalink] [raw]
Subject: Re: khttpd fate

On Fri, Jul 17, 2009 at 06:15:40PM -0700, Luis R. Rodriguez wrote:
> I was reviewing khttpd [1] history [2], and the last I see is it was
> merged for the 2.4 kernel with enthusiasm. I don't see any information
> about this for 2.6 though, nor can I find any other notes about why
> this was removed. Just curious if someone recalls why it was removed.
>
> Also, I really hate how trolly this questions sounds but here it goes anyway:
>
> Such userspace-kernel hacks shouldn't be necessary anymore based on
> 'performance/latency' arguments right? I take it khttpd wasn't serious
> but more of a hack for fun and now we should be able to laugh about
> it?
>
> [1] http://www.fenrus.demon.nl/
> [2] http://lwn.net/2001/0118/kernel.php3
>

I think it kind of got replaced by tux, which Red Hat shipped for a
while, but has been dropped now. I seem to recall davej mentioning a
while ago that apache had gotten much better at serving static content,
which is what khttpd/tux were very good at.

regards, Kyle

2009-07-22 00:40:01

by H. Peter Anvin

[permalink] [raw]
Subject: Re: khttpd fate

On 07/21/2009 05:20 PM, Kyle McMartin wrote:
>
> I think it kind of got replaced by tux, which Red Hat shipped for a
> while, but has been dropped now. I seem to recall davej mentioning a
> while ago that apache had gotten much better at serving static content,
> which is what khttpd/tux were very good at.
>

Also, lighttpd does really well, all in userspace. After all, static
http serving really is mostly a bit of header parsing followed by
sendfile(), so as long as a user-space process doesn't just sit on a
bunch of memory it can be done very cheaply.

-hpa

--
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel. I don't speak on their behalf.

2009-07-22 02:47:24

by Dave Jones

[permalink] [raw]
Subject: Re: khttpd fate

On Tue, Jul 21, 2009 at 05:36:25PM -0700, H. Peter Anvin wrote:
> On 07/21/2009 05:20 PM, Kyle McMartin wrote:
> >
> > I think it kind of got replaced by tux, which Red Hat shipped for a
> > while, but has been dropped now. I seem to recall davej mentioning a
> > while ago that apache had gotten much better at serving static content,
> > which is what khttpd/tux were very good at.
> >
>
> Also, lighttpd does really well, all in userspace. After all, static
> http serving really is mostly a bit of header parsing followed by
> sendfile(), so as long as a user-space process doesn't just sit on a
> bunch of memory it can be done very cheaply.

I think the rise of dynamically generated content was a big thing that
killed it off. With more and more of the web getting ajaxified, and the
php etc being offloaded to apache anyway, it just makes more sense to
have one webserver do everything as long as it's "fast enough".

I wrote something up on this a few years back when I made the decision to
drop Tux from the Fedora kernel. http://kernelslacker.livejournal.com/tag/tux

Dave

2009-07-22 04:52:36

by H. Peter Anvin

[permalink] [raw]
Subject: Re: khttpd fate

On 07/21/2009 06:57 PM, Dave Jones wrote:
> On Tue, Jul 21, 2009 at 05:36:25PM -0700, H. Peter Anvin wrote:
> > On 07/21/2009 05:20 PM, Kyle McMartin wrote:
> > >
> > > I think it kind of got replaced by tux, which Red Hat shipped for a
> > > while, but has been dropped now. I seem to recall davej mentioning a
> > > while ago that apache had gotten much better at serving static content,
> > > which is what khttpd/tux were very good at.
> > >
> >
> > Also, lighttpd does really well, all in userspace. After all, static
> > http serving really is mostly a bit of header parsing followed by
> > sendfile(), so as long as a user-space process doesn't just sit on a
> > bunch of memory it can be done very cheaply.
>
> I think the rise of dynamically generated content was a big thing that
> killed it off. With more and more of the web getting ajaxified, and the
> php etc being offloaded to apache anyway, it just makes more sense to
> have one webserver do everything as long as it's "fast enough".
>
> I wrote something up on this a few years back when I made the decision to
> drop Tux from the Fedora kernel. http://kernelslacker.livejournal.com/tag/tux
>

I can certainly explain why we don't use it on kernel.org, which is
almost all static content. We simply don't want to have to deal with
multiple web servers if we don't have to, and with sendfile() and
threading in Apache, it's reasonably efficient. If it wasn't, we would
probably go to lighttpd.

-hpa

--
H. Peter Anvin, Intel Open Source Technology Center
I work for Intel. I don't speak on their behalf.