httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Dean Gaudet <dgau...@arctic.org>
Subject Re: [STATUS] Fri Jun 6 08:12:55 EDT 1997
Date Fri, 06 Jun 1997 16:53:04 GMT
On Fri, 6 Jun 1997, Jim Jagielski wrote:
>  * Sleep

Check.

>  * Everyone have a drink or two of their choice

Check.

>  * More sleep

Check.

>  * Debate branching

I'm happy working on 1.3 right now, or something... the 2.x big changes
will happen, and when we get a patch on the block that says "big changes
ahead!" we can deal with the second repository then.  i.e. the first
patch which modifies the API.

>   * Workout path/goals for 2.0. Release 1.2.1 with below
>     fixes and improvements but let's give 1.2 some time
>     (and ourselves a rest).

Ok I'm going to humboldt for the weekend and I won't submit anything ;)

>   * Dean's solaris 256 FILE * problem
>     <Pine.LNX.3.95dg3.970507121700.11214J-100000@twinlark.arctic.org>
>     Status: Dean seeing if it works for user, maybe add to FAQ

Might as well expand this to:  consider the use of F_DUPFD to shuffle
descriptors around in order to avoid problems with third party libraries
and improve select() performance.  I'm thinking of something like this:

0 .. 2		the usual suspects
3 .. 31		free for per-request use	(class per_request)
32 .. 31+N	N listening sockets		(class selectable)
32+N and above	log files, and any other files	(class non_selectable)
		which modules want to open but
		not do selects on

Leaving 3 .. 31 open is to hopefully avoid any bugs in third party
libraries that get their underwear in a knot if they get a descriptor
above some magic K.  Note that the two classes requiring the use of
F_DUPFD are both determined at config time, so we're not incurring any
overhead during hits.

Putting the selectable sockets down lower saves time in select().

> Contrib stuff / future:

Whee!  So much of this :)

Dean


Mime
View raw message