httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Manoj Kasichainula <ma...@io.com>
Subject Why release 2.0 soon? (was Re: More detailed review of Ryan's filtering patch)
Date Mon, 24 Jul 2000 06:43:13 GMT
On Sun, Jul 23, 2000 at 06:37:40PM -0700, Rasmus Lerdorf wrote:
> I think there will be very little buy-in from the user community as a
> whole releasing 2.0 at this point.

My reasoning for saying 2.0 should go out the door is more for the
developers than the users. There's been talk of 2.0 since before I
got involved.

There are a few points during which there could or should have been
releases. Dean's Apache-NSPR work could have been an Apache "adjunct"
release potentially. Same with the single-process Apache-pthreads.
With some work (mostly on Windows), Apache 2.0 could have been the
multiprocess multithreaded server that was in pretty good shape last
year around this time. This is another such point that should be taken
advantage of.

> What could possibly motivate them to
> upgrade?  It is going to be less stable

Only because 1.3 was so insanely stable; 2.0 can't really be better,
except that signals should no longer be so troublesome, so there
should be fewer race conditions in theory.

> there is no PHP support

What's taking so long? :-)

> no mod_perl support

I thought this wass coming too. I think that making a concerted effort
to start closing down the release cycle now means that 2.0 gets
released around the end of the year; is that enough time?

> half the 3rd party UNIX libraries that they might attach
> via add-on modules are not thread safe.

There will be more of an impetus to make thread-safe versions of these
libraries when 2.0 is out, I'm sure.

> But if we release a 2.0 that doesn't provide a substantial number of
> end-user features above and beyond what they have with 1.3.x, and at
> the same time introduces all sorts of issues they need to deal with,
> they will at best try it out for a while and switch back to 1.3, and
> at worst not even bother installing it.

1.3 is doing the job for most people; that's why there are a lot of
servers running it. :) No matter what goes into 2.0 (other than the
User/Group vhost thing which actually feels hokey and painful to me;
think of keepalive and pipelining, yuck), I don't think the majority
of sites will upgrade for many months, and hopefully 3.0 will be out
first

> And no, I don't consider simply multi-threading to be a killer feature for
> the bulk of the users out there.  Yes, on certain platforms it will
> definitely be cool, but a threaded Apache on Linux is not going to be
> significantly faster although it should scale a bit better.  

Proper pth (or other userlevel threads) support shouldn't be hard to
put into Dexter; I think all that's needed is to eliminate the accept
serialization by using non-blocking accept. I'm afraid though that a
lot of the APRization done to dexter would have to be undone to make
this happen (which is one reason I didn't want dexter to be APRized in
the first place), but this change would essentially give us a more
bloated version of async I/O, which could improve scalability more
significantly, and could be added in time for 2.0.


Mime
View raw message