802.3ad and multiple mac addresses

There's nothing in the 802.3ad standard that says data traffic cannot use addresses different than that of the aggregator. All of that is handled up virtual interfaces sitting above the aggregator.

The only time that the aggregator's MAC address should come into play is when running L2 control protocols such as xSTP. In that case, an aggregator will have only one MAC address. In your example, the aggregator is simply forwarding traffic on behalf of virtual interfaces sitting above the aggregator.

Anoop

So , if I have multiple aggregators in team and each aggregator will have diff. mac address. and if I treat both the aggregators as active and pass traffic from Virtual interface on both the aggregators , what will happen on switch side ?

I mean

vip1/vmac1

| team | | (mac1)aggregator1 aggregator2(mac2)

So vmac1 will go on mac1 as well as mac2 So 2 virtual ports on switch side will see same mac address. Won't this confuse the switch ?

can anybody answer my question ?

This is getting away from the realm of 802.3ad, but...

If you use a single MAC address on the VIP you will confuse the switch. You would have to instantiate 2 VIPs one for each aggregator. The VIPs can then either differ in VLAN alone or in both VLAN and MAC address.

The situation you describe about is akin to having the same subnet configured on multiple ports in a NIC. I don't think they exist (but I could be wrong).

Anoop

It is possible on many platforms to configure different NICs (or

Yes. At least for all the multiport NIC's I've ever seen, each port has its own MAC address. Typically, consecutive MAC addresses. To the rest of the system, each port looks like its own NIC. (A little handwaving there, but not much).

_Is_ Sun still doing that? I thought they finally stopped doing that. (Thankfully, IMO).

rick jones

anoop wrote: (snip)

I have had multiple NIC on the same machine as different subnets on the same physical net.

Do multiport NICs have a different MAC address for each port?

Note that it is consistent with 802.3 for all ports (different NIC) on a machine to have the same MAC address, as is commonly done by Sun but not by most others.

-- glen

Actually, having the same MAC address on multiple MAC entities (i.e., multiple ports or NICs) is *NOT* consistent with the 802.3 architecture. It *is* consistent with the original Ethernet (DEC-Intel-Xerox) architecture. That is, 802.3 associates a MAC address with an interface; the "original Ethernet" associated a MAC address with a network station. Either model "works," as long as one is consistent.

-- 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

So it didn't get taken into 802.3 when other parts of DIX ethernet did?

I did once have an HP-UX machine running two subnets on the same physical net while we were waiting for some network hardware. After I did that, I realized that Suns wouldn't be able to do it. (Before secondary addresses or aliases were available. Though I did find that gated on HP-UX would route out a port with a different address with a metric 0 route.)

-- glen

FWIW, I always thought it was, let's say, "nearsighted," to allow one host with many interfaces to use a single MAC address. Unless I'm missing something, that idea only works if the multiple interfaces are each in different Layer 3 subnets, which is not something that one should assume at the outset will always be the case, IMO.

Functionally, I don't see that it makes a difference whether we're talking about DIX, 802.3, 802.4, 802.5, 802.11, FDDI, or even LANE, does it? The same problem occurs with any Link Layer protocol that depends on MAC addresses.

Bert

That is correct. It actually results from a subtle change in the formulation of the Pascal code that describes the MAC behavior.

The "one address per interface" model also lined up better with early PCs, where the base machine did not include any network interface, and NICs were sold as add-on peripherals. Since the NIC manufacturer had to provide an address for the NIC, multiple NICs would naturally have multiple addresses.

DEC and Xerox (and later, Sun) advocated the "one address per station" model in part because virtually every machine they sold was networked, thus it was convenient to provide the address as a function of the base machine.

-- 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

(snip)

When sending Sun boards in for service, one is supposed to remove the ROM, and replace it when the (possibly different) board is returned. We once had someone do this for a system, who then put the ROM (actually NVRAM) in backwards, destroying it. It took some time to get the address restored.

-- glen

In a normal network there is no reason to do it. Now with the widespread implementation of secondary (alias) addresses there is again not much reason for it.

-- glen