httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Bill Stoddard" <b...@wstoddard.com>
Subject Re: Suggested direction for fixing threaded mpm thread startup.
Date Mon, 16 Apr 2001 21:06:33 GMT
I don;t think we want to use a temp scoreboard if it is the same size as the 'normal' scoreboard.
 I
believe Dean pointed out that shared memory is not swappable. Maintaining a temp scoreboard
would
require twice the memory. Uuugh...

Bill
----- Original Message -----
From: Chuck Murcko <chuck@topsail.org>
To: <new-httpd@apache.org>
Sent: Monday, April 16, 2001 4:52 PM
Subject: Re: Suggested direction for fixing threaded mpm thread startup.


>
> On Monday, April 16, 2001, at 02:33 PM, <rbb@covalent.net> wrote:
>
> >
> >>>> What about the converse? Moving the old process/threads out of the
> >>>> scoreboard and into a "dying" temp area, and creating a new
> >>>> scoreboard
> >>>> entry for the old proc/thread entries.
> >>>>
> >>>> It seems safer (less potentially recursive, but a little slower).
> >>>
> >>> The problem with this, is that you take a very large chance that you
> >>> will
> >>> create more processes than you are supposed to.
> >>>
> >>
> >> Sorry I was thinking "overlaying the scoreboard" when I wrote "creating
> >> new scoreboard entry" and no the latter is not the way to do that. The
> >> former  minimizes the amount of scoreboard-type data being kept outside
> >> the scoreboard.
> >
> > The problem with overlaying the scoreboard, is that you have to
> > introduce
> > synchronization in order to do it.  You also need to modify how
> > processes
> > figure out where to store their scoreboard entries.  As soon as you
> > move a
> > scoreboard entry from it's default location to a new place in the
> > scoreboard, you are introducing a big headache.
> >
>
> Mmm, was thinking not of moving within the scoreboard, just copy out the
> few active threads with the old dying process and copy over all those
> entries for new.
>
> Must be synchronous, as you point out.
>
> Can your idea handle broken processes with a couple of hung threads? Is
> there a whole temp scoreboard, or just the overflow processes?
>
> Chuck Murcko
> Topsail Group
> http://www.topsail.org/
>


Mime
View raw message