Re: [mpls] Interim meeting: MNA implementation feedback

Hi all,

at the upcoming interim meeting, we will talk about our implementation 
experience with MNA in P4 on the Intel Tofino.

We will first briefly summarize the current state of the MNA framework 
relevant to our implementation and then present the challenges we faced 
and how we solved them. An abstract can be found below.

Throughout our presentation, we have some open points for discussion. We 
look forward to talking with you!

Title: Implementation Experience of MNA Using P4 on Intel Tofino Hardware

Abstract: In this talk we first summarize the MNA framework based on 
in-stack data (ISD) including requirements for backward compatibility. 
Only a few bits in the ancillary data (AD) are mutable for backward 
compatibility. In addition, multiple copies of hop-by-hop (hbh) network 
actions may need to be placed in the label stack to support intermediate 
nodes with limited readable label depth (RLD). This makes it difficult 
to carry mutable data in hbh network actions. Using post-stack data 
(PSD) for mutable data can solve these problems, but PSD come with other 
challenges. Either the entire stack must be parsed, or if this is not 
possible, the parser must bypass irrelevant parts of the header stack 
and parse the PSD instead.

We have built a prototype of the proposed MNA framework with P4 for the 
hardware-based switching ASIC Intel Tofino-1 running at 100 Gb/s. For 
PSD we assumed that the entire label stack fits into the RLD of the 
processing node. We explain the pipeline design of the implementation 
and derive additional constraints on the label stack. We propose 3 modes 
of operation for deploying the MNA framework in the network to highlight 
the tradeoffs in dealing with backward compatibility, RLD, and label 
stack constraints. These are a compatibility mode using only ISD, a 
compatibility mode using ISD and PSD, and a greenfield mode which is 
most efficient but not backward compatible. We implemented all 3 modes 
in P4 on the Intel Tofino, compared them in terms of implementation 
complexity, and discussed use cases. The presentation includes many 
examples for illustration, and the proposed deployment modes may trigger 
new discussions.

Best,

Fabian Ihle

On 09.04.24 18:08, Tony Li wrote:
>
> Hi all,
>
> We have scheduled a virtual interim meeting to discuss MNA 
> implementation feedback.
>
> Date: 2024-04-24
> Start Time: 08:00 America/Los_Angeles
> Duration: 01:00
> Remote Participation Information: 
> https://meetings.conf.meetecho.com/interim/?group=846d6161-d73f-4c7e-8f6f-a36961885373
>
> Thanks,
> Tony
>
>
> _______________________________________________
> mpls mailing list
> mpls@ietf.org
> https://www.ietf.org/mailman/listinfo/mpls

-- 
Fabian Ihle
Universität Tübingen
Fachbereich Informatik Lehrstuhl Kommunikationsnetze
Wilhelm-Schickard-Institut für Informatik
Sand 13, 72076 Tübingen

Raum: B303
Telefonnr.: +49 7071 29-70545
E-Mail:fabian.ihle@uni-tuebingen.de
Internet: uni-tuebingen.de

Re: [mpls] Interim meeting: MNA implementation feedback

Attachment: smime.p7s