From owner-freebsd-hackers  Mon Nov 11 14:15:55 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id OAA14694
          for hackers-outgoing; Mon, 11 Nov 1996 14:15:55 -0800 (PST)
Received: from brasil.moneng.mei.com (brasil.moneng.mei.com [151.186.109.160])
          by freefall.freebsd.org (8.7.5/8.7.3) with ESMTP id OAA14682
          for <hackers@FreeBSD.org>; Mon, 11 Nov 1996 14:15:47 -0800 (PST)
Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id QAA19941; Mon, 11 Nov 1996 16:12:15 -0600
From: Joe Greco <jgreco@brasil.moneng.mei.com>
Message-Id: <199611112212.QAA19941@brasil.moneng.mei.com>
Subject: Re: semaphores/shared memory
To: terry@lambert.org (Terry Lambert)
Date: Mon, 11 Nov 1996 16:12:14 -0600 (CST)
Cc: jgreco@brasil.moneng.mei.com, terry@lambert.org, scrappy@ki.net,
        twpierce@bio-3.bsd.uchicago.edu, hackers@FreeBSD.org
In-Reply-To: <199611112003.NAA18571@phaeton.artisoft.com> from "Terry Lambert" at Nov 11, 96 01:03:01 pm
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-hackers@FreeBSD.org
X-Loop: FreeBSD.org
Precedence: bulk

> > > However, now I have to question my assumptions... why is it necessary
> > > for the clients to signal the server?
> > 
> > Reuse of the buffer area?
> > 
> > It would be stupid for the server to start writing new data before
> > everyone else is done with it.
> 
> If 99/100 of the clients succeed, we should hang the other 99 for the
> one that is lagging out?

I guess that depends on the application.  :-)  I was not going to try
to dictate to someone else how to write a UNIX multi process application,
he already has the Stevens book.

> It would be useful to know if the data stream can be resynchronized,
> and what the actual effect of data loss for one client would be.
> 
> Also, the effect would be negligible if you double-buffered the
> client buffer area.  If the discrete buffer areas were large enough
> that the buffer could contain all the data sent in the maximum pool
> retention time window, then it would not be a problem.

Sure..  and sure again.  The question did not indicate that the data
was not already being double (or multi) buffered.  You would still need
to signal completion and acknowledgement on a per-buffer basis...  same
question, basically.

:-)

OTOH, a description of the problem being solved may not be a bad idea.
It will then let us make valid guesses as to how the process might be
improved...  ;-)

... JG