L2 vs L3 Etherchannel?

What is the real difference between L2 and L3 etherchannels? I can understand that with Layer 2 Etherchannels you can bundle two links into 1 big pipe, but why would I need to do this at Layer 3. Isn't it easier to just load share using routing over the multiple equal cost links?

Also, if you are doing Layer 2 etherchannels, do you have to configure the Po1 interface at all? I have seen people placing commands this interface such as "trunk encap dot1q" without actually making it a trunk ie using the "switchport mode trunk | dynamic" command.

And if you make it a layer 3 etherchannel, How is the IP addressing worked out. Do the switch interfaces have ip addresses in addition to the Po1 interface or is there only an IP address on the Po1 interface. If anyone has sample configurations used in a typical enterprise network, that would be greatly appreciated.

Thanks in advance.

Reply to
ciscortp
Loading thread data ...

I think that depends greatly on your configuration.

Think of the port-channel mainly as a layer 1 type of configuration in which you logically bind together two physical interfaces. Whether you make it a switched or routed interface occurs at another layer.

To me, port-channel seems more efficient than other means of load-balancing multiple interfaces. But I have encountered one gotcha. With the absence of keepalives between interfaces in port-channel, a medium in the middle, such as a Gig-E DWDM system, can maintain carrier but drop traffic resulting in a severely degraded network. At least with a routing protocol, keepalives will keep the session alive only if there is traffic flow. But if you are sure that carrier up means traffic flow, I would recommend port-channels hands down. Cisco may have implemented port-channel keepalives as I have not used these for over 2 years.

I do not know the answer to your last question. My guess is that the IPs on the individual physical interfaces do not matter because once it becomes part of the port-group, the interface is no longer directly addressable.

Best wishes Tim

Reply to
Tim F

load-balancing

use 802.1ad instead - the control protocol (LACP) tests the underlying links using keepalives. whether it can detect a link that trashes just some traffic is a different Q...

Cisco have implemented BFD (bidirectional forwarding detection) which is intended as a fast reacting connectivity check

l3 vs l2 is about the logical interface to the device, and also affects how the balancing algorithm decides which packets go on which link.

you can use L3 ECMP in combination with Etherchannel (or 802.1ad) - had to do this once to handle a 16 * GigE link.....

Dont forget you cannot do ECMP unless there is a L3 box at each end of the pipe (and that you can run a routing protocol that supports multi path, and that both boxes can support ECMP).

There are also some nice resilience protocols that allow a pair of boxes to "pretend" to support 2 links within a single load balanced 802.1ad layer 2 pipe - Nortel Passport 8600s have 1 version. The elegent bit is that anything that supports 802.1ad can sit on the end, but benefit from from the resilience of 2 links to separate central switches, without needing a routing protocol or any other complication.

cisco switches by default mainly support dynamic trunking protocol, which can sort out which ports go to the same destination switch and build the port group on the fly.

Frankly i think that anyone who depends on this kind of "feature" in a live network is just not paranoid enough....

DTP isnt going to work with other devices, such as other vendor switches, or even a conventional Cisco router

the IP address is bound to the set of links, not the individual interfaces.

Reply to
stephen

Think of the situation where the component links of an etherchannel on a switch are connected to a interfaces on a router. The router will have to treat those interfaces as a bundle since that is how the switch is treating them. Hence the router too will have to support etherchannel and have a corresponding interface that represents the bundle of component links. Since this is a router, that etherchannel interface will have an IP address and operate at layer3. And that is exactly what a layer3 etherchannel is.

There are some advantages to having a L3 etherchannel rather than having equal cost paths seen by L3 as two separate links. For instance, imagine that one of the links starts flapping. With an etherchannel, the routing protocol will not see that at all since the port-channel will always stay up. With independent equal-cost links rather than an etherchannel, the flapping links will cause routing updates and the attendant cpu activity trying to converge etc. In such a case, the etherchannel would give to load sharing and link redundancy with less overhead than if the links were treated as independent links.

No, the IP address will only be associated with the port-channel interface that represents the etherchannel and not on the component links. From the p.o.v of L3, the component links don't exist and only the port-channel interface is seen.

Cisco da Gama

formatting link

Reply to
ciscodagama

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.