Questions about UniDirectional Link Detection (UDLD) protocol

Pete89 · ‎10-08-2009

Hello,

Yesterday I was testing the failover of redundant links between a 3550 and a 4507 core switch.

I had created a etherchannel made up of two fastethernet ports and one gigabit fiber port on the 3550. I made a corresponding etherchannel on the 4507. I attached all ports mentioned and all was well. I did a show etherchannnel summary on the 3550 and I could see that the fiber port was marked P and the two fastethernet ports were marked s.

I then thought I would test the failover. I pulled the fiber cable from the 3550 and about thirty seconds later the fastthernet cables started forwarding traffic. So far so good.5 minutes Later I put the fiber cable back into the 3550. The link light went green right away and about 1 minute later the phone started ringing off the hook. The network was down. I pulled all the cables from the 3550 and in about 30 seconds the network was back to normal.

OK, this is what I think happened and this is where I would like your ideas too. I think what happened was the fiber link only went unidirectional and we got a STP loop.

If thats what happened I want to deploy UDLD on the fiber ports I have on my 3550s and on the core.

What I dont see is the purpose for normal mode UDLD is if it does nothing if it detects unidirectional traffic, and the aggressive mode scares me becuase I know one day UDLC will detect a unidirectional port which really isnt unidirectional and shut down the port. But since I have no experience with UDLC, I guess what I am asking is if this protocol has worked for you and is worth deploying.

So in my case can I use UDLC on ports in a etherchannel if I decide to go with it?

Many Thanks,

P.

Reza Sharifi · ‎10-08-2009

Pener,

When creating an etherchannel, the speed and duplex of all ports must be the same.

So, how did you create on etherchannel with mix of 1Gig and 100MB fiber interfaces?

Reza

Pete89 · ‎10-08-2009

Hi Reza,

Thanks for answering my post. I called this in to TAC and confirmed that Etherchannel will support different speeds as long as the port to port speeds are the same. In other words, my fastethernet ports on the 3550 were plugged directly into fastethernet ports on the 4507. I too thought that would be a problem, but I checked before doing it.

Abd like I said the etherchannel was funtional until I did my test.

Lucien Avramov · ‎10-08-2009

Check this doc:

http://www.cisco.com/en/US/tech/tk389/tk621/technologies_tech_note09186a008009477b.shtml

Pete89 · ‎10-08-2009

Hey Lavramov!

Thanks for the great link. It really does a good explanation of the protocol.

My only worries are what I said in my first post about the differences between the Normal and Aggressive Modes. Whats the tradeoff and benefits/drawbacks of either.

Thanks again,

P.

osnie · ‎10-08-2009

To reduce your problems, do it the best way. Don't mix speeds in an etherchannel. If you don't mix speeds in an etherchannel you will get subsecond convergence time and you will not get into any problem when bringing back your failed links.

Also, regardless if you're mixing speeds or not. Run PAgP or LACP since that will negotiate before bringing the port up, therefor you don't need UDLD.

Note. You can use the "errdisable recovery" command to bring back a port from an UDLD error.

Pete89 · ‎10-09-2009

Hi Osnie,

Thanks for your advice. I will take it into consideration, and so does my question then become:

Is mixing speeds in the etherchannel a BAD idea?

I use PAgP on all the links, and if what you say is true that PAgP should prevent unidirectional links from occuring, what happened to the core when the fiber link went back in?

Still I would like to hear what others think about mixing speeds on the etherchannel. Good? BAD?