Re: [dhcwg] Discussion on draft-ietf-dhc-failover-12.txt

In message <AE8D4698-C3E7-44B7-9201-B978D8024E0E@nominum.com>, Ted Lemon writes
:
> On Jan 24, 2005, at 5:46 PM, David W. Hankins wrote:
> > It seems we would at least like to know how much interest there was for
> > this within the WG still.
> I would very much like to see this draft advance, as I'm sure would a 
> number of other people.

I'm too new to using failover to make any detailed comments on the 
draft, but I want to echo this sentiment.  At our site, 
we're currently doing a long-overdue upgrade of our DHCP servers.  We 
need to use failover, and we'd REALLY like it to be a standard protocol.

> >     A2: "Eager update" instead of lazy update:
> >      [...]
> >         > Rephrased: is there enough deployments out there
> >         > with a fast connected primary-secondary to warrant
> >         > another failover protocol optimized for this?
> >
> >         Possibly.  I think many people read 'failover' and they
> >         dream of something different, something people would
> >         actually call 'High Availability'.
> >
> >         I think if you defined 'fast connected' as a direct
> >         crossover between the two, then no.  On the other hand,
> >         servers that are connected to the same switch are a dime
> >         a dozen.

They're a dime a dozen because it's easy to set up.  ("OK, we got these 
two servers; let's put 'em in that rack over there".)  This is fine for 
protecting against a server failure, but it does nothing for you in 
case of a *facility* failure.  We've had enough of those (power 
failures, flooding) that geographical separation between the DHCP 
servers is more important to us than a fast connection between them.

(Before you ask, we're also upgrading the facilities, so that flooding 
and power failures will be less likely).

> For the love of god, let's not redesign the failover protocol.   My 
> experience of this is that, considering the example of the ISC DHCP 
> server, there's no bottleneck in the protocol that's slowing the server 
> down.  The last thing you should be thinking about is redesigning the 
> protocol to accommodate the ISC server.   

Amen to that.  My understanding is that the failover protocol mostly 
works, and there's a lot of practical experience with it.  Let's move 
ahead with what we have now.

If people think that "eager updates" will help solve a problem, then they
can be explored as a possible future extension.

-- 
George C. Kaplan                            gckaplan@ack.berkeley.edu
Communication & Network Services            510-643-0496
University of California at Berkeley

_______________________________________________
dhcwg mailing list
dhcwg@ietf.org
https://www1.ietf.org/mailman/listinfo/dhcwg