[c-nsp] UDLD ?

[Jeff Fitzwater]

Thanks all for the info on UDLD.   In my case the test did not work as  
expected because the port was in auto-negotiate, as it should be.    
Disabling it allowed the port to stay up even if the other end was  
down (no light).
Enabling the UDLD worked as I would expect in this case.   But in the  
end the auto should be enabled and the UDLD will detect converter  
issues or patch panel issues.

The one thing I am curious about is, what happens if the other end is  
not CISCO but the next hop L2 is.  Does the UDLD packet (01-00-0C-CC- 
CC-CC) pass thru? ( I would say yes).  I would guess this could cause  
strange link failures and is why UDLD is not on by default.

The best reference for UDLD is the rfc 5171


Jeff



On Jun 30, 2009, at 11:25 AM, Tim Stevenson wrote:

> GE/10G can detect a physical unidirectional fiber link itself, UDLD  
> is not necessary to detect this type of failure.
>
> UDLD is needed for exactly the case you mention, or for cases where  
> one side of the link is "braindead" but does not bring the physical  
> link down (ie, software problem).
>
>
> HTH,
> Tim
>
>
> At 07:57 AM 6/30/2009, Peter Rathlev stated:
>
>> On Tue, 2009-06-30 at 09:59 -0400, Jeff Fitzwater wrote:
>> > We have had a few strange unidirectional link problems and I  
>> thought
>> > that I could detect them using UDLD.   So I thought I knew how it
>> > worked.   I
>> ...
>> > I thought that breaking on side of the fiber would only bring  
>> down one
>> > end LINK since the other still thought it was connected.
>> >
>> > I then disabled the UDLD and disconnect the fiber again and still  
>> had
>> > both ends show link failure.
>>
>> Just tried this between a 3560 12.2(35)SE5 and a 2970 12.2(25)SEC2  
>> with
>> the same symptoms as you describe; disconnecting one fiber doesn't
>> trigger UDLD but does give link down in both ends. This is also  
>> contrary
>> to what I expected.
>>
>> UDLD is useful in another case though: Media converters and EoMPLS
>> xconnected ports are transparent to UDLD but might not have link
>> poisoning enabled. With UDLD you would discover the loss of  
>> connectivity
>> to the neighbor even though the link doesn't go down.
>>
>> As long as the link actually goes down the UDLD isn't needed  
>> anyway. :-)
>>
>> Regards,
>> Peter
>>
>>
>>
>> _______________________________________________
>> cisco-nsp mailing list  cisco-nsp at puck.nether.net
>> https://puck.nether.net/mailman/listinfo/cisco-nsp
>> archive at http://puck.nether.net/pipermail/cisco-nsp/
>
>
>
> Tim Stevenson, tstevens at cisco.com
> Routing & Switching CCIE #5561
> Technical Marketing Engineer, Cisco Nexus 7000
> Cisco - http://www.cisco.com
> IP Phone: 408-526-6759
> ********************************************************
> The contents of this message may be *Cisco Confidential*
> and are intended for the specified recipients only.