[c-nsp] necessity of nowadays

[Phil Mayers]

On 23/03/16 15:21, Lukas Tribus wrote:
>> Bugs are completely in your control? You write your own network OS, and
>> build your own network hardware?
>
> I don't enable UDLD on my box and walk away before checking if it actually
> sees the link as bidirectional.

You seem to have taken my comment out of context. As I hoped the smiley 
indicated, I was being humorous. Sorry if that didn't come across.

> UDLD is extremly simple when compared to STP for example, so a quick test
> in the lab is not that hard.

If lab testing covered all real-world use-cases, all our jobs would be a 
whole lot easier ;o)

>
> Can you elaborate in which context UDLD false positived? New
> configuration/new links or suddenly after working fine for some
> time?

It's a number of years ago now, but we had a situation where UDLD 
decided *all* the links had gone unidirectional, and the device disabled 
them all. There was no apparent trigger.

It then failed to recover from the errdisable state, for no reason that 
we or TAC could determine.

Only our OOB access saved us.

We've had other issues where the cause was more obvious - downstream 
broadcast storm causing UDLD packets to fail to reach SUP, link gets 
dropped and the flap worsens the ongoing outage.

The sum total of our experience was "hasn't helped, has harmed".

I think in a lot of cases, alternative designs can completely sidestep 
the need for UDLD, and that is our preferred approach these days.

If UDLD is working for you, all power to you - your network, your rules.