httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Marc Slemko <ma...@worldgate.com>
Subject Re: cvs commit: apachen/src/main conf.h
Date Thu, 22 Jan 1998 00:15:54 GMT
On Wed, 21 Jan 1998, Marc Slemko wrote:

> I'm not sure if I would go for the "most others" thing when you get under
> high load.
> 
> The overhead of locking may more than cancel out the overhead of waking up
> all processes on accept().

I mean may be more than cancelled out by.

Probably not though.  

Chapter 2 of the new UNIX network programming has some numbers for various
methods of servers, eg. no accept() locking, accept locking, thread mutex
locking, descriptor passing, a copule of threaded models, etc.

> 
> Well.  I guess that depends.  Some OSes wake up all processes anyway when
> the lock is given to another process, some just pull one off the queue and
> leave the rest alone.
> 
> Note that on svr4 systems this is critical and you must have serialized
> accept no matter what. 
> 
> On Wed, 21 Jan 1998, Jim Jagielski wrote:
> 
> > jim@hyperreal.org wrote:
> > > 
> > > jim         98/01/21 15:12:09
> > > 
> > >   Modified:    src      CHANGES
> > >                src/main conf.h
> > >   Log:
> > >   A/UX can use SAFE_UNSERIALIZED_ACCEPT as well
> > >   
> > 
> > As Marc said, this is good for FreeBSD as well and, I'm guessing,
> > most others too (?). Maybe we should note what would cause it
> > NOT being a good idea somewhere in the docs and ask for
> > feedback from beta testers. Recall that flock mutex didn't work
> > until "recently" soooo that could be a good data point as well.
> > -- 
> > ====================================================================
> >       Jim Jagielski            |       jaguNET Access Services
> >      jim@jaguNET.com           |       http://www.jaguNET.com/
> >             "Look at me! I'm wearing a cardboard belt!"
> > 
> 


Mime
View raw message