[comp.protocols.tcp-ip] proxy arp

romkey@kaos.UUCP (John Romkey) (05/15/88)

Bill,

I hate to do this, but I don't like proxy ARP at all.

There are two reasons for why:

First, there are too many network media that don't use ARP for my
taste. ARPANET, X.25, ProNET-10. Maybe FDDI won't? I look at what you
said about these media this way: you're going to have to support
default gateways on them. You want the vendor to support it. The
vendor's IP layer should probably be pretty media-independent, so if
you've got default routers working in the software when you hook up a
system to the ARPANET, you've basically got what you need for ethernet
too. So it's not really an extra work to support it for networks other
than the non-ARP media.

Second, I had a really bad experience with an ethernet at MIT that had
a proxy ARP router on it a few years ago. Dave Bridgham was setting up
a second IP subnet on an ethernet at LCS. But a strange thing was
happening - his packets were disappearing. The proxy ARP router was
sending out ARP replies few milliseconds after the host Dave was
trying to talk to and it ended up gobbling up all the packets. It took
us a long time to track down the problem.

I think there are situations where you might want to set up a second,
test IP subnet or network on an network cable which already has a
different IP subnet or network on it, and that proxy ARP routers which
might believe they're doing exactly the right thing would make it
impossible to do this.

So proxy ARP is unacceptable to me.
-- 
			- john romkey
UUCP: romkey@kaos.uucp			ARPA: romkey@xx.lcs.mit.edu
 ...harvard!spdcc!kaos!romkey		Telephone: (617) 776-3121

philipp@LARRY.MCRCIM.MCGILL.EDU (Philip A. Prindeville) (05/17/88)

	I think there are situations where you might want to set up a second,
	test IP subnet or network on an network cable which already has a
	different IP subnet or network on it, and that proxy ARP routers which
	might believe they're doing exactly the right thing would make it
	impossible to do this.
	
	So proxy ARP is unacceptable to me.

But you should be able to configure a router to not know about a
network, and therefore not answer requests about it (to `unknow'
the network, as it were).  If you can't do this, it reflects a
flaw in the implementation, not the protocol.

-Philip

dab@ALLSPICE.LCS.MIT.EDU (05/17/88)

	Actually, there was one other factor that caused proxy-ARP to lose
in that case.  This was the use of a default route in the other gateway.
It didn't actually know a route to this test subnet I was setting up, it
just thought it did (via its default route).  So after proxy-ARPing the
address I was trying to get to, it forwarded the packet off to the ARPA
gateway (its default route), which forwarded the packet off to some random
core gateway (its default route) which hopefully dropped it before it
became an alligator in Dave Mill's swap.
						David Bridgham

romkey@kaos.UUCP (John Romkey) (05/19/88)

In article <8805170444.AA20271@Larry.McRCIM.McGill.EDU> philipp@LARRY.MCRCIM.MCGILL.EDU (Philip A. Prindeville) writes:
>But you should be able to configure a router to not know about a
>network, and therefore not answer requests about it (to `unknow'
>the network, as it were).  If you can't do this, it reflects a
>flaw in the implementation, not the protocol.
>
>-Philip


Suppose I don't own and run all the routers. Suppose the university or
corporate telecommunications office does, or suppose BBN runs one of
them, or the company down the street. I don't necessarily have control
over all the computers on my network, and the level of effort needed
to go through to get the necessary changes done to some of them for 15
minutes of testing may be prohibitive.
-- 
			- john romkey
UUCP: romkey@kaos.uucp			ARPA: romkey@xx.lcs.mit.edu
 ...harvard!spdcc!kaos!romkey		Telephone: (617) 776-3121

ralphw@ius3.ius.cs.cmu.edu (Ralph Hyre) (05/20/88)

In article <882@kaos.UUCP> romkey@kaos.UUCP (John Romkey) writes:
>
>Bill,
>
>I hate to do this, but I don't like proxy ARP at all.
...
>First, there are too many network media that don't use ARP for my
>taste. ARPANET, X.25, ProNET-10. Maybe FDDI won't? 
I suppose I should give Phil Karn a chance to say this first, but
AMPRnet (Amateur Packet Radio, network 44) generally won't - hidden
terminal problems and unreliable broadcast performance make it impractical.

Unfortunately, ARP has become too popular, so that people almost always think
of trying to use it to help with routing.
-- 
					- Ralph W. Hyre, Jr.

Internet: ralphw@ius2.cs.cmu.edu    Phone:(412)268-{2847,3275} CMU-{BUGS,DARK}
Amateur Packet Radio: N3FGW@W2XO, or c/o W3VC, CMU Radio Club, Pittsburgh, PA

karn@thumper.bellcore.com (Phil R. Karn) (05/21/88)

> I suppose I should give Phil Karn a chance to say this first, but
> AMPRnet (Amateur Packet Radio, network 44) generally won't - hidden
> terminal problems and unreliable broadcast performance make it impractical.

My code, written specifically for AMPRNET, does use ARP. We even have
our very own officially registered "hardware type" -- see the Assigned
Numbers RFC.

ARP on amateur packet radio works exactly like it does on Ethernet. Lost
ARP requests aren't a problem, since there'll be a retransmission (from
TCP or whatever) that simply gets turned into another ARP request. There
being no formal broadcast address in the AX.25 link layer protocol,
however, we had to define our own -- "QST". (You hams out there will
understand the significance of these letters :-)).

The only complication comes when "digipeaters" are used. These are
simple store-and-forward repeaters that use a source routing feature in
the link protocol.  Broadcasting through digipeaters doesn't work, so
you have to manually enter the proper source route and destination
address into your ARP table.

Phil

bdale@hpcsla.HP.COM (Bdale Garbee) (05/28/88)

/ srg@quick.COM (Spencer Garrett) /  2:15 am  May 18, 1988 /

>Why can't a host just ARP for any destination and expect the
>appropriate gateway(s) to answer?  If there were a hopcount
>field in the ARP record, I think this would solve all the problems.

We (N3CVL, K3MC, and myself) looked very hard about a year ago at 
doing something like this in the KA9Q
Internet Package, to improve the situation on amateur packet radio, where
almost noone can talk directly to anyone else, and so gateway/switch issues
become very important.

I eventually backed away from this idea when it became more clear to me that
a distinction can and perhaps should be maintained between "routing" which
is a logical operation, and "address resolution" which is (to me at least) a
purely physical operation.

However, my recent experience in designing and maintaining a sitewide LAN
at work, including a bunch of discless clusters that for reasons of performance
and cost will be subnetted by putting two ports on each discless server, have
led me to be very interested in Proxy ARP, and other "ARP extensions".  (If only
HP would officially support Proxy ARP in HP-UX... sigh)

My curiosity is up, therefore I also would be interested in comments on 
this subject.

Bdale, N3EUA

rcb@cccvs1.ncsu.edu (Randy Buckland) (02/01/89)

Does anybody know where I can get a copy of proxy arp code for either
an IBM PC/RT V2.2.1 or an Ultrix system V3.0?

Randy Buckland
rcb@ncsuvx.ncsu.edu

meggers@mothra.nts.uci.edu (Mark Eggers) (09/01/90)

Is there anywhere I can get proxy arp that will run on a Sun 4 ??

thanks - /mde/