The result of using a Cisco Gigastack GBIC depends on the cabling setup, you either have a point-to-point connection (in which case EtherChannel across switches still won’t work) or a cascaded stack connection (where a cross-switch EtherChannel might be possible). To be honest I haven’t used the Gigastack GBIC in that way before, bu maybe this article @ cisco.com will shed some light on the Gigastack configurations:

Catalyst Switch GigaStack Configuration and Implications

Basically, if your switches are indeed stacked, once you log in to one switch that switch should report all the ports (so also the ones on the other physical switches in the stack) as its own. So for two stacked 48-port switches you should for example see ports GigabitEthernet1/1 through 1/48 and GigabitEthernet2/1 through 2/48 (or something along those lines, module numbers may differ).

I hope this helps. If you need any further info I’m here always. Try the veal

In the end, I really want the failover more than the speed. If I can’t have both, I’m okay with leaving my setup as is. Obviously I’ll look into a 4 port NIC for my next servers. Thanks again!

If you’re using teaming software on the NIC most of the time the teaming software will use one (virtual) MAC address as the source address, and without some intervention STP will indeed disable one of the ports. The reason for this is simply that Cisco does not support creating EtherChannels across multiple switches, and thus STP will detect the same MAC address on two different ports in the STP tree, assume that a loop has occurred and block one of them.

There’s two ways around this: either set the teaming mode on the NICs to something like “failover only”, or disable STP on the switch ports. I normally strongly advise against the latter option, as STP is your friend when it comes to accidental problems in the wiring of your network.
So if you want the dual link for redundancy, set the teaming mode to “failover only” (or something like that, it differs per vendor), if you want the dual link for speed, connect both links to the same switch.

Of course there is a third way: use a 4-port NIC (or 4 teamed 1-port NICs, or two teamed 2-port NICs), connect 2 ports to switch A in one EtherChannel and 2 ports to switch B in one EtherChannel. This way you’ll have 2 ports running on one switch using the full bandwidth of both ports and 2 ports on the second switch for when the first switch or the connections fail.

Thank you for this article. It’s exactly what I was looking for. I’m teaming a 2 port Intel NIC via 802.3ad. I’d like one port to go to my “A” switch and the other port to go to my “B” switch. Switch A & B are Cisco 4506 switches (IOS 12.2), interconnected. According to your post, STP will disable one of my two ports, correct? Is there a way to configure both ports to remain active and load balance through both switches?

You can use normal straight cables for any switch-to-server connections, including bundled EtherChannel links.

http://www.cisco.com/en/US/tech/tk389/tk213/technologies_tech_note09186a0080094714.shtml#matrix

You’d normally use a destination-based method on the server, and a source-based method on the switch it’s connected to. I’d normally use a MAC-address based method over an IP-address based method (but that’s personal I guess).

This way all outgoing sessions to different clients a server initiates will be spread across the available links (because it has one source (itself) and multiple destinations (the clients), and all incoming sessions from clients to the server will be distributed by the switch across the available ports.

I hope this answers your question. If not, I’m here