BILLW@MATHOM.CISCO.COM (William Westfield) (07/23/87)
I heard an interesting rumor that the IBM distributed TCP/IP source routes all its packets. Is this possibly true???? BillW -------
JBVB@AI.AI.MIT.EDU ("James B. VanBokkelen") (07/25/87)
I heard an interesting rumor that the IBM distributed TCP/IP
source routes all its packets.
Is this possibly true????
BillW
-------
What I have been told is this:
1. IBM's TCP/IP for the PC, as released for use on their Token Ring Adapter,
does not ARP in the conventional sense: Instead, it sends the ARP as an
"all rings" broadcast, with an empty Routing Information Field. When the
reply comes back, it looks for the filled-in RIF field, and uses it. No
RIF field, no function.
2. IBM's TCP/IP for AIX can behave as above, but only if you enable it
with some magic switch.
3. Other IBM TCP/IP products will try the "all rings" approach, but only
if they get no reply to a conventional ARP.
4. IBM bridges don't pass conventional (not "all rings") broadcasts.
So, yes, folks, source routing is alive & well at Yorktown, whether or not
ISO has accepted it. No official position is being stated here, but comments
I've heard range from "That's WRONG! Let's try to talk them into conforming."
to "Oh, hell, they won't listen, and they're going to sell piles of that
junk - It's gonna be a dirty job, but we'd better change the code..."
I've heard nothing about aberrant behavior on Ether, so apparently they're
only punishing their own pioneering customers at the moment....
jbvbDCP@QUABBIN.SCRC.SYMBOLICS.COM (David C. Plummer) (07/29/87)
Date: Fri, 24 Jul 87 23:32:20 EDT
From: "James B. VanBokkelen" <JBVB@AI.AI.MIT.EDU>
I've heard nothing about aberrant behavior on Ether, so apparently they're
only punishing their own pioneering customers at the moment....
Maybe they're just trying to go one up on DEC? Remember that DEC
violates the spirit of the Ethernet hardware address by forcing the
Ethernet address to be algorithmically determined by the DECnet address,
resulting in Ethernet addresses which do not indicate the vendor of
hardware interface, as well as not ensuring globally unique addresses.sandrock@uxc.cso.uiuc.edu (07/31/87)
>/* Written 1:30 pm Jul 29, 1987 by DCP@QUABBIN.SCRC.SYMBOLICS.COM in uxc.cso.uiuc.edu:comp.protocols.tcp-ip */ > > Date: Fri, 24 Jul 87 23:32:20 EDT > From: "James B. VanBokkelen" <JBVB@AI.AI.MIT.EDU> > > I've heard nothing about aberrant behavior on Ether, so apparently they're > only punishing their own pioneering customers at the moment.... > > Maybe they're just trying to go one up on DEC? Remember that DEC > violates the spirit of the Ethernet hardware address by forcing the > Ethernet address to be algorithmically determined by the DECnet address, > resulting in Ethernet addresses which do not indicate the vendor of > hardware interface, as well as not ensuring globally unique addresses. > DECnet uses the node address to set the least significant 16 bits of the 48-bit Ethernet hardware address while setting the most significant 32 bits to a "known" constant value. Specifically, the Ethernet address will be AA-00-04-00-xx-xx, where the xx-xx fields are the DECnet node address (area-number * 1024) + node-number. There may be both certain advantages and also disadvantages to this approach, but is it true that these addresses are not globally unique? Mark Sandrock, (sandrock@uxc.cso.uiuc.edu)
DCP@QUABBIN.SCRC.SYMBOLICS.COM (David C. Plummer) (08/03/87)
Date: 31 Jul 87 16:30:00 GMT
From: uxc.cso.uiuc.edu!sandrock@a.cs.uiuc.edu
DECnet uses the node address to set the least significant 16 bits of the
48-bit Ethernet hardware address while setting the most significant 32
bits to a "known" constant value. Specifically, the Ethernet address
will be AA-00-04-00-xx-xx, where the xx-xx fields are the DECnet node
address (area-number * 1024) + node-number.
There may be both certain advantages and also disadvantages to this
approach, but is it true that these addresses are not globally unique?
Mark Sandrock, (sandrock@uxc.cso.uiuc.edu)
The DECnet scheme "only" allows 64K hosts, or probably more precisly,
only 64 areas. I would be quite surprised if there are only 64
installations of DECnet in the world. Therefore, if we assume that
people conventionally number node within an area starting with 1, then
there is likely a global non-uniqueness between machines of two
installations that have the same area number. I have no idea how DECnet
area numbers are assigned.
The major disadvantage is that if ACME Computers (read: some other
vendor) also used an algorithmic approach, it would be impossible, 100%
impossible, for a machine to be able to support both protocols at once
on one transciever, since it would require the transceiver to have two
Ethernet addresses. Such multi-protocol machines do exist. Luckily,
there is only one vendor that I know of (DEC) that uses an algorithmic
Ethernet address. I'm pretty sure DEC knew about ARP in time; I'd have
to get some ancient mail files off of tape to make sure.haas%gr@CS.UTAH.EDU (Walt Haas) (08/03/87)
In article <171500006@uxc.cso.uiuc.edu>, sandrock@uxc.cso.uiuc.edu writes: > DECnet uses the node address to set the least significant 16 bits of the > 48-bit Ethernet hardware address while setting the most significant 32 > bits to a "known" constant value. Specifically, the Ethernet address > will be AA-00-04-00-xx-xx, where the xx-xx fields are the DECnet node > address (area-number * 1024) + node-number. > There may be both certain advantages and also disadvantages to this > approach, but is it true that these addresses are not globally unique? > > Mark Sandrock, (sandrock@uxc.cso.uiuc.edu) Not globally unique by a long shot - there can be only 65534 DECnet addresses by this number system. However there is an even more embarrassing problem with this arrangement- (and this problem also occurs with XNS which does the same thing). Suppose you have a DECnet (XNS) host which is connected to two Ethernets. Now each DECnet (XNS) node is allowed only a single DECnet (XNS) address, which the software loads into [all of] the Ethernet interface[s] attached to the node. This means that the node has the same 48-bit Ethernet address on both Ethernets. OK, now what happens when you attach a learning Ethernet bridge, such as the DEC LANbridge 100 or any of the numerous competitors between the same pair of Ethernets? Well, what happens is that when the node sends a packet on one Ethernet, the bridge learns the node is on that net, and starts routing all packets with that destination to the net where it saw the node's address. Then the same node sends a packet on the other net, using the same Ethernet source address, and the bridge says "whoops, that node just moved" and commences to route all packets for the node across to the other Ethernet! ----------------* Cheers -- Walt ARPA: haas@cs.utah.edu uucp: ...utah-cs!haas "...The unforgivable but by no means uncommited sin in this connection occurs when, A calling down that his rope is caught, it devolves that B is himself standing upon it. Any second man once guilty of this precious bit deserves to be demoted from his position for the remainder of the season!" Robert L. M. Underhill, "ON THE USE AND MANAGEMENT OF THE ROPE IN ROCK WORK," Sierra Club Bulletin, February 1931, p. 80.
sy.Ken@CU20B.COLUMBIA.EDU (Ken Rossman) (08/03/87)
We've had a reasonable amount of experience with medium to large DECnet networks here at Columbia. We have our own internal network, and we also connect to various other sites who also have their own DECnet networks. I am not sure how Ethernet addresses are administered (I was under the impression that one or more of the larger corporations plugging Ethernet divvy up the board addresses by board manufacturer). In any case, the first four bytes of a DECnet transformed Ethernet address are, by some "global allocation" method, preassigned to DEC as I understand it. Since, as David Plummer points out, DECnet IV only currently supports 64K nodes on any single DECnet (well, really 63K --- area 0 is not a real area, and generally designates Phase III nodes), it doesn't really matter whether the limitation occurs at the ethernet addressing level or at the DECnet addressing level. The limit is still the same -- 63K. There is no "global" administration of DECnet areas that I know of. I deal them out for CCnet (the DECnet here that spans the CU campus and several other schools). In fact, awhile back, we were looking at connecting to other large DECnets (PHYSnet, I think, was the name of one of them, or something close to that), and we were given AN area to live in by the PHYSnet administrators. Since we were already multi-area, this made it basically impossible for us to connect with them, and we bagged the idea. As for DECnet, ARP, and ethernet addresses, Ultrix handles this just fine. We just make sure that DECnet comes up before IP does, so that the ethernet address that ARP uses is the DECnet-transformed one. /Ken -------
DCP@QUABBIN.SCRC.SYMBOLICS.COM (David C. Plummer) (08/04/87)
Date: Mon 3 Aug 87 16:54:35-EDT
From: Ken Rossman <sy.Ken@CU20B.COLUMBIA.EDU>
We've had a reasonable amount of experience with medium to large DECnet
networks here at Columbia. We have our own internal network, and we also
connect to various other sites who also have their own DECnet networks.
I am not sure how Ethernet addresses are administered (I was under the
impression that one or more of the larger corporations plugging Ethernet
divvy up the board addresses by board manufacturer). In any case, the
first four bytes of a DECnet transformed Ethernet address are, by some
"global allocation" method, preassigned to DEC as I understand it.
Sorry. Ethernet addresses are, in theory, assigned to hardware
manufacturers in 2^24 address chunks and the manufacturer is responsible
for administering its 2^24 addresses. DEC is "reassigning" the original
hardware Ethernet address to be a DEC-specific protocol-related hardware
address. That is not the intention of the Blue Book. DEC does not
"manufacture" Interlan, 3Com, Symbolics, TI, etc, hardware, yet the
hardware Ethernet addresses used would indicate DEC does.
...
As for DECnet, ARP, and ethernet addresses, Ultrix handles this just fine.
We just make sure that DECnet comes up before IP does, so that the ethernet
address that ARP uses is the DECnet-transformed one. /Ken
That's not what I meant. What I meant was that ARP existed in time for
DECnet IV to use it, and that if DECnet used ARP instead of its current
algorithmic translation, (a) we could preserve globally unique
addresses, (b) you wouldn't have to worry about bringing up DECnet
before IP (or any other protocol), and (c) if somebody else goes against
the intention of the Blue Book we could still run DECnet and that other
protocol at the same time (since it wouldn't then require different
Ethernet addresses).paclark%ford-cos1.arpa@FORD-COS1.ARPA (Patricia A. Clark) (08/04/87)
sandrock@UXC.CSO.UIUC.EDU (Mark Sandrock) (08/05/87)
We don't have a DECnet host 1.1, and neither should any site which has links to the "outside world". All DECnet addresses on our campus are assigned by the PHYSnet management, which includes international address assignments as well. Also, I doubt that if "ACME" vendor tries to market a network that conflicts with DECnet that they will have many sales to systems which already have DECnet installed. They would most likely (and deservedly) soon be a defunct company.
sandrock@UXC.CSO.UIUC.EDU (Mark Sandrock) (08/05/87)
Thanks for your message re: DECnet and Ethernet. All of our DECnet addresses here at the University of Illinois are assigned by the PHYSnet management (whoever that is). We have not yet seen the need for multiple areas on this campus. Also, one would expect that DEC might anticipate the need and expand the DECnet addressing scheme to something reasonable in a future version (Phase V?).
sandrock@UXC.CSO.UIUC.EDU (Mark Sandrock) (08/05/87)
It sounds like you are claiming that DEC preempts Ethernet hardware addresses which have been officially assigned to other vendors, but I have yet to see a specific example of such an address. Is it asking to much either to see an example or else have everyone drop this claim?
sandrock@uxc.cso.uiuc.edu (08/05/87)
I hope you will bear with me... not being yet familiar with the workings
of UNIX mail I don't know if my responses to individual mail will also
be posted here, but I would like to clarify my statements re: DECnet...
I still have not seen any example of a DECnet-assigned Ethernet address
which is NOT globally unique. Can someone give such an example of DEC's
preemption of another vendor's Ethernet address?
I realize that it would be possible for 2 DECnet nodes to choose the
same DECnet address thereby choosing the same Ethernet address, but in
practice of course this is not supposed to happen, anymore than 2 sites
choosing the same IP address.
While I also do not know of any "global administration" of DECnet ad-
dresses, it so happens that here at the University of Illinois we make
all DECnet address assignments through a contact with the PHYSnet man-
agement (no, I don't know any more details than that).
To say this another way, I am free to bring up my own Ethernet (in my
office, right!) and choose any DECnet and/or IP addresses I want for my
personal machines, but once I connect my net to the Internet or PHYSnet
or whatever, I obviously must choose to coexist in harmony (or suffer
the consequences!).
Also, I agree that 2**16 unique addresses is an undesirable limitation,
but at one time it was only 2**10 addresses, and so one might surmise
that DEC will once again respond to the perceived need for more addresses
in due time.
This may all be a moot discussion in light of newer protocols being
developed (?), but then who can really say? (Not me anyway (:^))
Mark Sandrock, (sandrock@uiucuxc.UUCP)hedrick@TOPAZ.RUTGERS.EDU (Charles Hedrick) (08/06/87)
As far as I know, in phase V the DECnet address will consist of two bytes of subnet and then the Ethernet address. Once phase IV compatibility is not needed (typically this would be at the next phase), one could presumably stop using the special Ethernet address forced on you by the phase IV DECnet design, and use the preassigned Ethernet address. So at that point, machines using DECnet would have unique addresses. I think this discussion has diverged from its intent. I was simply trying to explain what was meant when someone said that machines running DECnet did not have globally unique Ethernet addresses. Your comments about how DECnet addresses are assigned on your campus show that there can't be any conflict in DECnet addresses at your installation. However the fact remains that your hosts no longer have globally unique Ethernet addresses. There are almost certainly at least two hosts on the DEC corporate network with the same address (since they have several DECnet networks each of which have very full address space), and presumably some on other large corporate networks as well. This is not a criticism: obviously your machine will not become confused with those other machines. It is simply an explanation of a statement in a previous posting.
DCP@QUABBIN.SCRC.SYMBOLICS.COM (David C. Plummer) (08/06/87)
Date: Wed, 5 Aug 87 11:05:49 CDT
From: sandrock@uxc.cso.uiuc.edu (Mark Sandrock)
Also, I doubt that if "ACME" vendor tries to market a network that
conflicts with DECnet that they will have many sales to systems which
already have DECnet installed. They would most likely (and deservedly)
soon be a defunct company.
Well, what if it were UniSYS, IBM or GM? Muscle flexing is a very
impolite thing to do anyway, and in open network environments it may
achieve a marketing objective at the expense of thwarting the kinds of
atmospheres that lead to experimentation and creativity.
Date: Wed, 5 Aug 87 11:18:21 CDT
From: sandrock@uxc.cso.uiuc.edu (Mark Sandrock)
Thanks for your message re: DECnet and Ethernet. All of our DECnet
addresses here at the University of Illinois are assigned by the
PHYSnet management (whoever that is). We have not yet seen the need
for multiple areas on this campus. Also, one would expect that DEC
might anticipate the need and expand the DECnet addressing scheme to
something reasonable in a future version (Phase V?).
So Hedrick believes (Phase V). Unfortunately, it sounds like they are
going off in another wrong direction. From Hedrick's description, they
will in the future be basing DECnet protocol addresses on the Ethernet
hardware address. Well, what happens if your Ethernet hardware breaks
and you need to replace it? Answer as far as I can tell: you have to
force the new Ethernet address to be the same as the old. What happens
if you recommision the old board on the same net? Another problem: Does
this mean that DECnet Phase V has variable length addresses? What do
they do for networks that have a different number of hardware address
bytes than the Ethernet has? My belief is that protocol addresses are
logically attached to the machine as an entity, that hardware network
addresses are attached to hardware interfaces, and that they should not
be related because of the current problems with (1) DECnet Phase IV and
(2) my understand of Hedrick's description of Phase V. IP, Chaos and
PUP (and probably others I don't know about) do it right.
Date: Wed, 5 Aug 87 11:24:18 CDT
From: sandrock@uxc.cso.uiuc.edu (Mark Sandrock)
It sounds like you are claiming that DEC preempts Ethernet hardware
addresses which have been officially assigned to other vendors, but
I have yet to see a specific example of such an address. Is it asking
to much either to see an example or else have everyone drop this claim?
Sure. There is a Symbolics 3600 at our site that has an Ethernet
hardware address of 08-00-05-03-00-38. The 08-00-05- portion was
assigned to Symbolics by the Ethernet number Czar for use by Symbolics
for its Ethernet interfaces. That is the "official" hardware address of
the interface on that machine. That machine also has a DNA address of
41.69. Because of the way DNA works, the booting procedure for the
machine changes the Ethernet address of the interface to
AA-00-04-00-45-A4. Therefore, DNA has preempted our officially assigned
address.SYSTEM@CRNLNS.BITNET (08/06/87)
Mark, > It sounds like you are claiming that DEC preempts Ethernet hardware > addresses which have been officially assigned to other vendors, but > I have yet to see a specific example of such an address. Is it asking > to much either to see an example or else have everyone drop this claim? Herewith the DECnet Ethernet address for node 19.53, aka 19509:: AA-00-03-01-10-E7 I do not know if this conflicts with any other vendor's address assignments. Selden E. Ball, Jr. (Wilson Lab's network and system manager) Cornell University NYNEX: +1-607-255-0688 Laboratory of Nuclear Studies BITNET: SYSTEM@CRNLNS Wilson Synchrotron Lab ARPA: SYSTEM%CRNLNS.BITNET@WISCVM.WISC.EDU Judd Falls & Dryden Road PHYSnet/HEPnet/SPAN: Ithaca, NY, USA 14853 LNS61::SYSTEM = 44283::SYSTEM (node 43.251)
hedrick@TOPAZ.RUTGERS.EDU (Charles Hedrick) (08/06/87)
No, DEC does not preempt other vendors' addresses. Let's try this again. The Ethernet spec requires that a vendor allocate each Ethernet interface an address which is globally unique. That is, there is no other interface in the world with the same number. DEC interfaces have such an address. However when you turn on DECnet, the processor causes the interface to change addresses. It no longer uses its globally unique address, but instead uses an address that is produced algorithmically from the DECnet address. The first 4 bytes of the address are a constant. The last two are the DECnet address with various bit twiddling. (The exact algorithm is documented in the DECnet manual.) This address is obviously unique within your site, and even within the set of sites that are connected by DECnet, but is no longer globally unique, because any other site with the same DECnet address will now have the same Ethernet address. The Ethernet address is otherwise legal. The vendor prefix is DEC's. However the address as a whole is not unique across the whole world. Perhaps the confusion is whether DECnet addresses are globally unique? A DECnet address has two parts: area and host. There are 64 possible area numbers and 1024 possible host numbers. This is not a large enough address space to allow DECnet addresses to be globally unique. Again, they are obviously unique for any given DECnet network. But your DECnet network, DEC's, GM's, etc., may and probably do use the same addresses. The physics net, in which both of us participate, controls allocation of area numbers for the schools that participate. So among that set of schools DECnet addresses are unique. But there can only be 62 schools in that network, because the network itself uses one area number, and 0 is not allowable. There are certainly more than 62 universities in the country. Indeed there was a message on dcom.lan not long ago from someone who couldn't join the network because their local DECnet needed several areas, and therefore their address allocations conflicted with the physics nets'. There are other large organizations, such as DEC's engineering network, that use most of the possible DECnet addresses. So the conclusion is: DECnet addresses are unique within the specific DECnet network, but not on a world-wide basis. Furthermore, the address space is small enough that there could not be a DECnet network as large as the IP internet. Simple arithmetic will show that. DECnet uses 16 bits for its address. In general 16 bit addresses do not provide enough address space to allow globally unique addresses. Thus the 16-bit networks, including chaos, PUP, and DECnet, are used primarily as local networks, or between specific sets of institutions. For a large network, at least 32 bits is needed, and even that may be marginal. IP uses 32 bits, DECnet phase V uses 64 bits, and XNS uses 96 bits. All of these are designed to be used with large-scale networks.
haque@umn-cs.UUCP (Samudra E. Haque) (08/08/87)
> Herewith the DECnet Ethernet address for node 19.53, aka 19509:: > > AA-00-03-01-10-E7 > > I do not know if this conflicts with any other vendor's address > assignments. > This message possibly has got nothing to do with the actual subject.. But is there ANY gateway from the INTERNET to DECNET (dec's private network). Specifically I am interested in contacting an individual who works in the Sunnyvale Digital ?Plant? ?office?. He has often said that there is no way that we can get messages accross the network walls. Is there anybody out there willing to give me a route to his node? If you mail me a note I will mail you back what information he gave me and will try your suggestion. I am not that terribly familiar of VMS based networking. Samudra E. Haque Computer Science Systems Group (CSSG) Computer Science Department University of Minnesota, Minneapolis (612) 625-0876 haque@umn-cs.ARPA , haque@umn.UUCP
sy.Ken@CU20B.COLUMBIA.EDU (Ken Rossman) (08/11/87)
Herewith the DECnet Ethernet address for node 19.53, aka 19509::
AA-00-03-01-10-E7
I'm afraid I fail to see where you get AA-00-03-01-10-E7 from 19.53.
First of all, the DECnet-transformed ethernet address should start with
AA-00-04-00. Secondly, the last two octets of the DECnet/ethernet address
is the 16 bit DECnet node number, which consists of 6 high order bits of
area, and 10 low order bits of node number. The bytes are swapped also.
So in the above example, if you take the last two bytes, 10 and E7, swap
them to get E710, and then break them down to 6 and 10 bit groups, you get:
111001 1100010000
(area) (node num)
or, node 57.784.
This ethernet address was probably still the original hardware address
before DECnet got ahold of it and mangled it. /Ken
-------