On Thu, Jan 3, 2013 at 8:01 PM, Tony Li <[email protected]> wrote:
> Example please?  Assuming that the prefixes are parsed out of the message 
> correctly, then treating them as withdrawn should be hazard free.

If a transit router in your AS treats-as-withdraw a prefix, other
routers may try to send traffic across it, but the router which has
treated-as-withdraw will either have no information for that route, or
it will have made a different bestpath choice.  This could cause a
loop or a blackhole.

> Non-sequitur.  Cyclic resets are orthogonal to treat-as-withdraw.  As 
> enumerated above, ignore-bad-message is harder to implement and is more 
> dangerous.

No, ignore-bad-message is not harder to implement.  To say so is simply a lie.

To treat-as-withdraw you have to follow some rules, parse a damaged
message to look for NLRI, and then withdraw them.

Treat-as-withdraw is unquestionably harder than simply ignoring the
whole contents of a message.

Your argument is some hand-waving about how hard it is to ignore a
message, and your incorrect belief that treat-as-withdraw can't leave
the RIB in state that can cause forwarding problems.

I'm not picking at your position just for the sake of argument.  I
agree with you that error-handling (current draft) is too complicated.
 I'm suggesting alternatives, as well as modifications that should be
made to either RFC4760bis or to error-handling to make it actually
work.

Just to move beyond ignore-bad-message vs. treat-as-withdraw arguments
for now, I believe some basic questions need to be answered for
treat-as-withdraw to be a useful option.

1) should a BGP speaker be able to promise that MP attributes will be
sent first?
   if so, a capability code is needed
   this could be done by error-handling or RFC4760bis

2) should non-MP NLRIs in the message be moved in the way
error-handling suggests?
   if so, a capability is absolutely required
   the basic encoding of BGP Messages changes if this is to be done

3) there are numerous attribute-specific rules in error-handling now.
They are not crazy but they are complicated.  Is this complexity
enough to deter vendors from implementing error-handling at all?  Are
they likely to catalyze bugs?  Do the attribute-specific rules do more
harm than good?

If you are in favor of treat-as-withdraw then the above are important
questions to answer.

-- 
Jeff S Wheeler <[email protected]>
Sr Network Operator  /  Innovative Network Concepts
_______________________________________________
GROW mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/grow

Reply via email to