On 4/4/2016 5:28 PM, bruno.decra...@orange.com wrote:
My issue is how do we prove that_nobody_ is using it? Proving the negative is
hard, and silence is not part of the proof. To prove the negative, we would
need explicit statement from everyone, which looks impossible.
I think you're exaggerating a little ;-) I'm having trouble believing
that you are unable to determine whether the 3107 "multiple labels"
feature is in use in any of your company's deployments.
I think a more serious concern is how to avoid tickling the "day 1" bugs
that may exist. There may be implementations, supporting only a single
label, that fail to set the S bit. There may be implementations,
supporting only a single label, that fail to check the S bit. These
"day 1" bugs will never have caused a problem, because the multiple
labels feature has not been used. We should endeavor to ensure that
these"day 1" bugs do not cause a problem in the future if newer
implementations begin to use the multiple labels feature.
Right now, we're arguing about which of the following two strategies is
more likely to cause a problem:
1. In the first strategy, 3107bis requires the S bit to be set when
there is a single label. This will allow 3107bis to interoperate with
3107 implementations of "multiple labels", but it will not allow 3107bis
to interoperate with (buggy) 3107 implementations that send a single
label, but don't set the S bit.
2. In the second strategy, 3107bis assumes, in the absence of the
Capability, that there is only a single label, and doesn't bother to
check the S bit. This will allow 3107bis to interoperate with (buggy)
implementations that send a single label but fail to set the S bit; it
will not allow 3107bis to interoperate with (non-buggy) 3107
implementations of multiple labels.
My argument is that the second strategy is better because it will be
less disruptive. This is based on my belief that the "day 1" bugs do
exist, and that the "multiple labels" feature has yet to be deployed.
Your argument seems to be that the first strategy is better because (a)
it only causes disruption if the 3107 implementation has a bug, in which
case the bug can be fixed, and (b) if the second strategy is used, a
3107-compliant implementation of multiple labels will fail to
interoperate with a 3107bis-compliant implementation of multiple labels,
and both implementors will claim compliance.
I think your argument is reasonable, the question is really just which
strategy will cause less disruption.
Do other members of the WGs have opinions about this?
___
BESS mailing list
BESS@ietf.org
https://www.ietf.org/mailman/listinfo/bess