Re: [mpls] What percentage will carry indicator/ancillary data in the future

[Loa Andersson <[e-mail]>]

Kireeti,

I had an off-line very short chat with Stewart on this, and what I took 
away from that discussion is that we probably need to drive the on what 
percentage of MPLS packet that carry indicators + ancillary data based 
on use cases.

I don't think the term ancillary is ambiguous, we have been over it 
several time in the DT and we have agreed every time.

Ancillary data are data data that helps the node take the correct 
forwarding decision. There are three types of ancillary data (if you 
look at how they are positioned in the packet:
- post stack data (PSD)
- in stack data (ISD)
- No data (NoD)

Counting NoD with ISD and PSD as ancillary data, is a bit of a tweak, 
but I think it is good from a descriptive point of view, and also there 
is some implicit data folded in with the indicator.



On 11/10/2021 18:18, Kireeti Kompella wrote:
> Hi Loa,
> 
> The term ancillary data is ambiguous: as I understand, it refers to both ISD and PSD.
> 
> Two points:
> 1. High value traffic will frequently use FAI/ISD/PSD.  Best effort traffic probably will probably do so less often.

   - I tned to agree, but it is kind of unproven and would require real
     measurements to prove

> 2. As we find more use cases for FAI, the percentages for both will go up.

   - this is likely
> 
> For example, mandating CW for L2 PWs (as has been done, and may be reinforced) will mean more post-stack headers.

   - agreed
> 
> The way I understand your question is: how often is the PSD looked at 

Well, yes and no, looking at PSD is part of it and might cost more 
processing than other ancillary data.
(since it potentially creates lots of work for the forwarding engine).

Yes. And that was where I started to think about this.

We should consider the answer to be: very often (for high-value 
traffic), and make sure we handle this case well.  Even if the average 
number is small (lots of best effort traffic with no PSD), it wouldn’t 
serve the community well to only drop a few packets, but all of them 
high-value :)

Yes, I thnk that is a good strategy.

/Loa

> 
> :k
> 
>> On Oct 9, 2021, at 10:43, Loa Andersson <[e-mail]> wrote:
>>
>> Design Team,
>>
>> I have one thing that I been thinking about.
>>
>> If we consider future mpls encapsulated packet, what percentage will carry indicators and ancillary data?
>>
>> /Loa
>> -- 
>>
>> Loa Andersson                        email: [e-mail]
>> Senior MPLS Expert                          [e-mail]
>> Bronze Dragon Consulting             phone: [phone]
>>
>> _______________________________________________
>> mpls mailing list
>> [e-mail]
>> https://www.ietf.org/mailman/listinfo/mpls

-- 

Loa Andersson                        email: [e-mail]
Senior MPLS Expert                          [e-mail]
Bronze Dragon Consulting             phone: [phone]