Re: TWAMP analysis for assisting BFD debuggin (was Re: BFD stability follow-up from IETF-91)

[Jeffrey Haas <[e-mail]>]

Greg,

On Mon, Dec 22, 2014 at 06:46:26AM +0000, Gregory Mirsky wrote:
> if you assume that BFD flap is more likely because of implementation or
> overloading system rather than because of network condition, then I???d
> like to understand how intermittent congestion in a source end-node can be
> differentiated from intermittent congestion in the network without PM.

Congestion is one of the scenarios covered by tracking things at the BFD
level.  Another is simply whether packets are being generated or not or even
the simple timings within protocol packet generation.

While very lightweight, BFD still has implementation scaling choices that
can lead to unusual behaviors.  Packets are inherently jittered within the
context of the protocol, but to some extent it's simply a practical
consideration that maintaining tight timings is difficult.  Similarly, if a
packet simply is not sent because an internal timer ran late and it decided
to send the next packet instead, that's a consideration.

Being able to tie BFD issues to either issues within implementation vs.
issues happening on the wire is useful, but the tool sets will overlap in
some cases.

> I???m not questioning value of improved debugability of an implementation,
> BFD in this case, but I???d prefer it not to have adverse impact on the
> protocol, on its ability to perform its main task most efficiently.
> Debugging, I believe, been and still is, proprietary, implementation
> specific. Problems you???ve described, I believe, unlikely to be
> experienced in HW based BFD implementation. 

Unlikely to be experienced in HW?  I suspect you'll get a much different set
of feedback off-list where people think they can speak a bit more bluntly
than on-list.

> Thus, what benefit, changes
> you propose, would bring for such implementations?

Even in a situation where a given implementation was perfect, we care about
any pair of implementations.  When services flap due to BFD, you now have
two devices to troubleshoot and blame.

-- Jeff