EtherChannel and STP

[alexmchau]

I just started studying BCMSN and read about EtherChannel. I keep thinking why we still use STP when there are EtherChannels. My stupid question is :-

If Ethernet Channel can utilizes both (or more) links for improved speed and also provide an automatic fail-safe benefit in case one of the links fails, then we can design a network with all Ethernet Channel links with redundancy benefits without using STP at all. True or False?

Thanks. - Alex

[Jon Marshall]

Alex

Not a stupid question at all.

One thing to remember is that except for the 6500 VSS you cannot run an etherchannel link across switches. So lets say you have 2 distribution switches DS1 & DS2 which are connected via an etherchannel trunk. All switch interconnects are layer 2 etherchannel trunks.

Now you want to connect an access-layer switch (AS1) to the distribution switches using etherchannel trunks. You could just connect it to DS1 with an etherchannel and this would protect against individual link failures within the etherchannel.

But what if DS1 fails ?. Your access-layer switch now has no path to the distribution layer. So you need to connect your access-layer switch to both switches. And as soon as you do this you have a loop ie.

AS1 -> DS2 -> DS2 -> AS1.

So you need STP to block one of the access-layer etherchannel uplinks.

Note that due to the performance of layer 3 switches these days you can design a network utilising only L3 links between switches which negates the need for STP across your switch interconnects but as with everything there are pros and cons to this approach.

Jon

[Jon Marshall]

Sorry, that should read

AS1 -> DS1 -> DS2 -> AS2

Jon

[Jon Marshall]

Okay note to myself. Do not try and answer questions after a nights drinking !!

That should read

AS1 -> DS1 -> DS2 -> AS1

Jon

[Joseph W. Doherty]

If the L2 links between individual switches (or where multiple switches at least appear as one switch, e.g. 3750/3750-E stacks or 6500 VSS), and their multiple interconnecting links are bonded together such as Etherchannel, and there are no physical loops, then STP isn't necessary. However, often still a good idea to avoid accidental physical looping of your L2 domain.

PS:

BTW: Using Etherchannel you will obtain link redundancy, but might lack other redundancy that might be provided by STP.

E.g. chassis dual Etherchannel with termination on the same card; lose the card you lose the link.

Also, Etherchannel has the potential for providing more bandwidth than STP which blocks redundant paths, but a Etherchannel doesn't automatically convey additional bandwidth. Single flows don't use multiple physical links of the Etherchannel and multiple flows don't always either. (The latter depends on how the flow maps into the hash algorithm of the Etherchannel being used.)

[alexmchau]

Thanks you all for your explanation. That helps a lot !!!

- Alex