Ascend Archive
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: (ASCEND) Arp routing



> On Wed, Apr 08, 1998 at 11:19:01AM -0400, Dave Van Allen wrote:
> > There's no such thing as "arp routing", what you have here is Tomfoolery.
> 
> While I would also not call it ARP routing, I don't at all see why some
> people bash it or dislike it.

Because it doesn't scale, and because (having not been designed for this
use) it does not include the necesary features. See below.

> > arp was not designed to be used as a routing or re-direction function across
> > different subnets. Proxy-arp in a situation like this is a bad hack and
> 
> It was not designed for that, but it was designed flexible enough to allow
> for that. It allows for even more esoteric hacks like "ARP-for-everything"
> which is euphemistically called "IP Switching" or "Layer 3 Switching".

It is flexible enough to allow a box like a Max to proxy for addresses
which are supposed to be on the local LAN but are actually in its pool.
Other similar uses are possible as well. It is not flexible enough to
handle the situation that was asked about in the original mail where
the [ MAC <--> IP ] correspondence may change over time.

You allude to the technique which Pipelines are able to use, if so
configured, to reply to ARP requests for itself, plus anything at all
that is not on the local LAN. The assumption is that the station is
sending ARP requests for non local addresses because it was neglected to
put a default route in it.

As far as I can tell, this hack is just for the purpose of avoiding the
need to put a default route into every station. But it is not required.
The IP protocol suite provides a router discovery mechanism as a part of
its ICMP protocol. The ICMP router discovery also very nicely handles
load balancing and backup situations, where you may have a router that
goes away. It can be (and is) implemented on hosts (that is, machines
which are not expected to be running any routing software). It is
lightweight so there is no argument not to use it.

Using ARP for this hack? Explosive ARP tables (each host is going to end
up with one ARP entry for every host it contacts, whether the host is
close or far), no handling of load balancing or backup.

(Note about load balancing: A host does not HAVE to react to multiple
router discovery announcements with the same preference by load
balancing. The point is only that it CAN).

If you are a host which happens to be sending packets to the wrong
router, ICMP has also provided a mechanism which can and should be
used to rectify that: ICMP redirects.

ARP provides no such mechanism. Packets are going to have to bounce off
the incorrect router possibly indefinitely.

Even worse, networks that use ARP heavily are *VERY* hard to debug. I
know I've always had trouble finding an elusive ARP problem.

-Phil
++ Ascend Users Mailing List ++
To unsubscribe:	send unsubscribe to ascend-users-request@bungi.com
To get FAQ'd:	<http://www.nealis.net/ascend/faq>


References: