Re: [mpls] [Bier] WGLC: draft-ietf-bier-mpls-encapsulation-06

[Eric C Rosen <[e-mail]>]

On 5/8/2017 11:44 AM, Stewart Bryant wrote:
> The document seems to be about how you operate BIER in an MPLS network, 

Please avoid the word "operate".  We don't want to give the 
Operations&Maintenance ADs any opportunity to insist that the document 
is not complete without a "how to" guide for operators ;-)

> and so I would have expected tunnelling over MPLS to be in scope. 

The topic of the draft is how to form a BIER encapsulation that can be 
used in an MPLS network, and how to form a BIER encapsulation that can 
be used in a non-MPLS network.  Admitedly, there is some discussion 
(sections 2.1.3 and 2.2.3) of further encapsulating the BIER packets in 
tunnels, but not a comprehensive survey of all the issues that might 
arise when tunneling.

Suppose P1 and P2 are packets, and E(Px) is the result of encapsulating 
Px in order to send it through a particular tunnel.  I think it is true 
in general that if P1 and P2 have the same entropy, then it is desirable 
for E(P1) and E(P2) to also have the same entropy.   This isn't 
mentioned because it isn't specific to either BIER or to MPLS, it's a 
general principle of tunneling.  But I could add something like the 
above sentences to sections 2.1.3 and 2.2.3, as well as to section 6.9 
of the architecture draft.  Note that that text deliberately avoids 
RFC2119 language.

I don't want to get into any detail about how to set the entropy in the 
tunnel encapsulation headers; that's clearly out of scope. There is no 
need for a discussion of when it is and is not appropriate to use the 
entropy label, how one knows whether the entropy label will be 
understood properly, how one knows whether the entropy label will make 
the stack too long, whether the FAT PW mechanism is appropriate, whether 
hardware can always preserve the entropy properly when putting on a 
tunnel encapsulation header, what the best entropy mechanism is for 
MPLSoUDP encapsulations, etc., etc., etc.

Someone could write an entire draft on the entropy issues that arise 
when tunneling, but I don't think this needs to be dealt with in the 
BIER docs.

(BTW, I'm sure you've noticed that the entropy field in the BIER header 
is 20 bits long, and hence could be copied into an entropy label.  
Perhaps it's worth mentioning tht explicitly, as long as RFC2119 
language is not used.)