need for redundancy: ethernet-address twice in one LAN

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Hello everybody,

this may sound like a strange thing but it does make sense in my case:

Context:
I want 2 machines to receive the same ethernet packages (for redundancy
reasons). The communication is unidirectional and uses UDP on top.
Currently this works by having 2 nodes with the same HW-/Ethernet-address
and IP-address on one hub(!)

Question:
How can I do the same thing with a switch?
Originally I thought this wasn't possible, then I found out about port
mirroring which kinda-works (but won't scale). Later on, somebody
told me that this (having the same HW-address on 2 switch-ports) is
supposed to work out of the box with ANY switch. Can anybody help me on
this last statement? Of course I tried that and it wouldn't work, it just
sent data the the port that last received the answer to an ARP.
Any hints?

Thanks a lot!
-Rainer

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <pan.2005.01.11.13.27.16.848043@birkenmaier.org>,
Rainer Birkenmaier <rbrk001@birkenmaier.org> wrote:

> Context:
> I want 2 machines to receive the same ethernet packages (for redundancy
> reasons). The communication is unidirectional and uses UDP on top.
> Currently this works by having 2 nodes with the same HW-/Ethernet-address
> and IP-address on one hub(!)
>
> Question:
> How can I do the same thing with a switch?
> Originally I thought this wasn't possible, then I found out about port
> mirroring which kinda-works (but won't scale). Later on, somebody
> told me that this (having the same HW-address on 2 switch-ports) is
> supposed to work out of the box with ANY switch. Can anybody help me on
> this last statement? Of course I tried that and it wouldn't work, it just
> sent data the the port that last received the answer to an ARP.
> Any hints?
>

The switch should forward frames whose DA is *unknown* (to the switch)
out all ports other than the one on which it arrived. Once your devices
start sending ARP messages, the switch will "learn" the address, and
will then forward only to the port on which it last heard the ARP.

One "solution" is to make sure that your receiving devices never send
any frames (or at least, never send any frames using the "duplicated"
address); the switch will then never learn the address and continue to
"flood" frames to all ports. (You did say that this was a receive-only
application; if it *truly* is, you should be able to turn off all
transmissions.)

A better "solution" was suggested by another poster: Assign a multicast
address to your application, and enable that multicast address on the
two (or more) machines that you want to hear these frames. This is
*precisely* the sort of thing that multicast was designed for.


--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <pan.2005.01.11.13.27.16.848043@birkenmaier.org>,
Rainer Birkenmaier <rbrk001@birkenmaier.org> writes:
>Hello everybody,
>
>this may sound like a strange thing but it does make sense in my case:
>
>Context:
>I want 2 machines to receive the same ethernet packages (for redundancy
>reasons). The communication is unidirectional and uses UDP on top.
>Currently this works by having 2 nodes with the same HW-/Ethernet-address
>and IP-address on one hub(!)
>
>Question:
>How can I do the same thing with a switch?
>Originally I thought this wasn't possible, then I found out about port
>mirroring which kinda-works (but won't scale). Later on, somebody
>told me that this (having the same HW-address on 2 switch-ports) is
>supposed to work out of the box with ANY switch. Can anybody help me on
>this last statement? Of course I tried that and it wouldn't work, it just
>sent data the the port that last received the answer to an ARP.
>Any hints?

Yes. Stay with the hub. It is the perfect device, as it replicates
a packet entering port "A" to all ports "not A". Trying to convert
a switch into a buffered repeater is at best a hack if it works at
all.
Thus said in case of GE where there are no hubs available one will
have to bind the application to a multicast address.
Being a polite person, I am not going the comment on the "strange"
application...

--
Manfred Kwiatkowski kwiatkowski@zrz.tu-berlin.de

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <pan.2005.01.11.13.27.16.848043@birkenmaier.org>,
Rainer Birkenmaier <rbrk001@birkenmaier.org> wrote:
:this may sound like a strange thing but it does make sense in my case:

:Context:
:I want 2 machines to receive the same ethernet packages (for redundancy
:reasons). The communication is unidirectional and uses UDP on top.
:Currently this works by having 2 nodes with the same HW-/Ethernet-address
:and IP-address on one hub(!)

There are probably other ways to achieve the redundancy, but
I lack the experience in that area. Depending on what has to be
accomplished in your situation, you might find the material
on Vincent C. Jone's web site to be useful.


:Question:
:How can I do the same thing with a switch?
:Originally I thought this wasn't possible, then I found out about port
:mirroring which kinda-works (but won't scale). Later on, somebody
:told me that this (having the same HW-address on 2 switch-ports) is
:supposed to work out of the box with ANY switch. Can anybody help me on
:this last statement?

It's obviously false. Many switches do not allow you to assign
particular MAC addresses to ports. The MAC address of a particular
port isn't even necessarily allocated by a fixed algorithm -- it
could be dependant on the order the ports came up. Static MAC
allocation for layer 2 ports seem to be quite common, though.

Something I read awhile ago indicated that each layer 2 port is
required to have distinct MAC addresses in order for spanning tree
to work properly. I have not investigated this point to determine
whether there are allowed exceptions beyond turning off spanning
tree for those ports.

:Of course I tried that and it wouldn't work, it just
:sent data the the port that last received the answer to an ARP.
:Any hints?

Oddly enough, just a couple of hours ago, I happened upon something
that would perhaps suit your needs. I was reading through the
release notes on the Cisco 3750 series, for IOS 12.1(14)EA or so
[might have been 12.2(14)SE] and saw a "caveat" described there.
It seems in that particular software release, if you had
assigned a static MAC address to a port with active port security,
then packets destined for that MAC were flooded to every
port in the VLAN. {I'm probably missing out on a detail or two
of what was required to cause the situation.} The implication
is that 1) the Cisco 3750 allows user-specified MAC addresses
on ports; and 2) you might be able to turn the flooding bug
to your advantage in your situation. You could probably find
the bug description by googling cisco.com for
3750 release-notes flooded

There may be other methods of achieving the desired end.
Most modern switches allow per-vlan spanning tree and per-vlan
MAC address tables so as to handle situations in which the same MAC
address might show up on different ports of a switch due to
participation in different VLANs or due to redundant links.
Possibly HSRP would be suitable for your needs.

Sorry for the vagueness: I've only had -useful- modern multilayer
switch for a couple of weeks and there is a lot to learn on it.
--
Rome was built one paycheck at a time. -- Walter Roberson