[c-nsp] necessity of nowadays

[Lukas Tribus]

> Hi List,
>
> i've been pondering about the real need for udld nowadays, each time it
> bites me in a case of false positive. At least since we have gigabit SFPs it
> became almost impossible to willfully provoke an unidirectinal link: The
> physical port allready detects missing light and goes down.

In case you have a different control plane protocol that covers the same
exact failure scenario (like LACP), sure, go for it.

But it has to be implemented the same way, and not all features, like RFI
or auto negotiation, cover all cases that UDLD or LACP do.

A unidirectional link can have lots of reasons and physical link problems are
the smallest problem.

About false positives: I run UDLD across basically all catalyst and most ME
platforms and false positives occur only with long obsolete software
on long obsolete platforms or due to erroneous configuration. And
in my experience they happen when you enable UDLD or the link,
not some time later in production.


On the list we have strong opinions against UDLD.

The reason I am pro UDLD is because:

- a had meltdowns because UDLD was not enabled
- a had issues where UDLD prevented a meltdown succesfully
- UDLD never caused a single production issue (yes, it false positiv'ed sometimes,
but due to wrong configurations or bugs, both of which are completely in my
control and don't happen sporadically at 04:00 in the moring, while hardware
does fail at 04 in the morning)


Clearly people made very different experiences with UDLD, but I wouldn't
run a STP network with out it (I wouldn't run STP networks at all, if I would
have a choice, of course).


Lukas