Re: [mpls] draft-fang-mpls-label-forwarding-no-swap - how much does it really save?

[Andrew Qu <[e-mail]>]

Sasha,

My comments are in lines

From: Alexander Vainshtein [mailto:[e-mail]]
Sent: Friday, July 24, 2015 10:31 AM
To: Andrew Qu; Robert Raszuk; Shahram Davari
Cc: [e-mail]
Subject: Re: [mpls] draft-fang-mpls-label-forwarding-no-swap - how much does it really save?


Andrew,

Inline below.


Thumb typed on my LG,
Sasha

------ Original message ------
From: Andrew Qu
Date: 24/07/2015 11:07
To: Alexander Vainshtein;Robert Raszuk;Shahram Davari;
Cc: [e-mail]<mailto:[e-mail]>;
Subject:RE: [mpls] draft-fang-mpls-label-forwarding-no-swap - how much does it really save?
Hi Sasha,


My comments in lines…




From: mpls [mailto:[e-mail]] On Behalf Of Alexander Vainshtein
Sent: Friday, July 24, 2015 7:26 AM
To: Robert Raszuk; Shahram Davari
Cc: [e-mail]<mailto:[e-mail]>
Subject: Re: [mpls] draft-fang-mpls-label-forwarding-no-swap - how much does it really save?


Hi all,

Please note that CONTINUE in SPRING is:

1. A DP-agnostic primitive

2. *Implemented* as SWAP instruction with MPLS DP.



[Andrew]  You are contradicting with 1) and 2).

[[Sasha]] SPRING is supposed to work over IPv6 DP as well.

CONTINUE over this DP simply means retaining the same extension header and forwarding according to nirmal IPv6 processing,



[Andrew] it does not matter,  by saying 1) , claiming 2) is not necessary.

Now that  you did 2) as fact,  then why adding    “NO_SWAP” is so bad?









With “CONTINUE” defined in SR draft,  for NEW ASIC to be designed,

Why should I still use “SWAP” to achieve it?

[[ Sasha]]Because the draft in question explicitly defines that CONTINUE is implemented as SWAP?

The SPRING drafts mention that it could be SWAP to the same label, but no more than that. They do NOT require NO-SWAP anywhere.





[Andrew] again,  MOST important is that SR introduce “CONTINUE” as concept.  Whoever

                Read the doc,  NO_OP is natural impression.  SWAP with same value is “HACKING” not the solution to me.

                I am surprised that in MPLS DP,  NO_SWAP is inferior to SWAP from concept point of view.







Considering that I might

Have good saving in precious on-board memory for the ASIC ?



[[Sasha]]As alreadyvdiscussed on this thread, this is a local implementation matter that does not require any standartization.





[Andrew] Why  you think SWAP is NOT local implementation but NO_SWAP IS ? as I don’t understand why we MUST NOT

                Avoid mentioning both “SWAP and NO_SWAP for CONTINUE for user to chose.





Cheers! and thanks for the comments



Andrew