From owner-freebsd-isp  Wed Oct  2 09:04:00 1996
Return-Path: owner-isp
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.5/8.7.3) id JAA17714
          for isp-outgoing; Wed, 2 Oct 1996 09:04:00 -0700 (PDT)
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 JAA17706
          for <freebsd-isp@freebsd.org>; Wed, 2 Oct 1996 09:03:56 -0700 (PDT)
Received: (from jgreco@localhost) by brasil.moneng.mei.com (8.7.Beta.1/8.7.Beta.1) id LAA05494; Wed, 2 Oct 1996 11:02:15 -0500
From: Joe Greco <jgreco@brasil.moneng.mei.com>
Message-Id: <199610021602.LAA05494@brasil.moneng.mei.com>
Subject: Re: RAID Controller Product
To: james@blacksun.reef.com (James Buszard-Welcher)
Date: Wed, 2 Oct 1996 11:02:15 -0500 (CDT)
Cc: jgreco@brasil.moneng.mei.com, james@blacksun.reef.com, cassy@loop.com,
        freebsd-isp@freebsd.org
In-Reply-To: <9610020847.ZM9039@blacksun.reef.com> from "James Buszard-Welcher" at Oct 2, 96 08:47:12 am
X-Mailer: ELM [version 2.4 PL24]
Content-Type: text
Sender: owner-isp@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

> OK, this is now officially a DNS question...
> 
> On Oct 2,  8:23am, Joe Greco wrote:
> > Subject: Re: RAID Controller Product
> > > Wouldn't there be a delay for clients still trying to reach
> > > your news server? If they have cached an IP address for
> > > news.wherever.com, and then you took it out of Round Robin,
> > > would there still be a finite number of clients trying to
> > > reach that IP address?  (Assuming they aren't looking to your
> > > nameserver and you didn't HUP it).
> > > 
> > > I'm pretty sure that Netscape doesn't (or at least didn't with
> > > 2.0) query the nameserver each time...
> > 
> > Netscape's loss, not mine.  If they do not honour my TTL, that is
> > their own freaking problem.
> 
> 	Gotcha. But for the length of your TTL, would there
> 	be some of your clients going to the wrong IP address?
> 	The one that's down?

In a crash?  That is acceptable.  If I have a TTL of one minute, it is
also not too big a deal.  They will get a connection time out, and then
the next time they try, round robin will cause them to pick a different
machine.

During a downtime?  Remove the machine from the round robin pool in
advance.  Simple fix.

> > Question:
> > 
> > Would you rather have your service entirely unavailable because
> > something strange happened and your box panicked and locked up?
> > Because some malicious soul hacked their way in and decided to
> > newfs your root filesystem?  Etc.?
> > 
> > I would rather have total redundancy :-)
> > 
> > ... JG
> >-- End of excerpt from Joe Greco
> 
> 
> 	I'm with ya. I would rather have total redundancy.
> 	But it seems like there would be a period where you
> 	*didn't* have total redundancy because some clients
> 	would still hit the bad IP address because their
> 	local-nameserver (for instance) has cached RR A
> 	record... do you just lower your TTL to a small number?

You still have total redundancy.  You just do not necessarily have
100% guaranteed connection attempts.  But as far as I am concerned,
if I have a crash and people can not connect every 1 out of N times
(where N >= 2) then I am better off than if I have a crash and people
can not connect every 1 out of 1 times.

So you do everything you can to minimize the chance of them
connecting to a dead address.

... JG