Hi Haoyu,
Thank you for your review and comments.
Although I'm in sympathy with your concerns to some extent, the network
operators are seeking for a PFC like solution in WAN, mainly for the AIDC
interconnection scenario.Please see inline for some more responses.
Original
From: HaoyuSong <[email protected]>
To: 肖敏10093570;[email protected] <[email protected]>;
Date: 2026年02月28日 08:17
Subject: RE: [rtgwg] Congestion Notification for Pause
Hi Xiao Min,
Thank you for submitting the draft. Below are some quick comments after I read
it.
PFC aims to achieve lossless ethernet within DCN, so I guess the propagated
back pressure pause packet is also intended for lossless WAN. But the WAN has
much longer path span and the latency could be tens to hundreds of
milliseconds. The buffer headroom required could be primitively large.
[XM]>>> As far as I know, there are already some routers equipped with large
buffer, and I'm sure more is coming.
PFC is applied for a whole priority queue on a link which could include many IP
flows from many paths. Sending pause packets for each of them could have a
large overhead and bandwidth impact.
[XM]>>> Yes, you're correct. One potential mitigation is for the egress PE to
select a few larger IP flows and send pause packets for the selected IP flows
only.
PFC has many negative side effects such of HoL blocking, PFC propagation,
potential deadlock. That’s why it’s only used as the last resort when e2e
congestion control can’t make it. Extending it to WAN could only exacerbate
the situation. The pause could paralyze the entire WAN. Therefore, I think the
mechanism should be meticulously evaluated and tested before considering the
standardization.
[XM]>>> As far as I know, the PFC is widely deployed in AIDC although it has
many shortcomings as you indicated. The realities show its value. The solution
in this draft was designed with PFC's shortcomings in mind, if there is
alternative proposal, I'll be glad to discuss and compare. Anyway, IMHO
employing PFC like pause mechanism in WAN is the right direction.
Cheers,
Xiao Min
Cheers,
Haoyu
From: [email protected] <[email protected]>
Sent: Friday, February 27, 2026 12:19 AM
To: [email protected]
Subject: [rtgwg] Congestion Notification for Pause
Dear all,
A new document draft-xiao-rtgwg-congestion-notification-for-pause-01 has been
submitted. Link as below.
https://datatracker.ietf.org/doc/html/draft-xiao-rtgwg-congestion-notification-for-pause-01
This draft defines the IP Pause message which is analogous to Ethernet PFC
frame.
Except for the different layers, the IP Pause message is distinct from the PFC
frame in that, the IP Pause messages are all sent from one node (the egress PE)
while the PFC frames may be sent by multiple nodes along the path with stepwise
backpressure.
Looking forward to your review and comments.
Cheers,
Xiao Min
_______________________________________________
rtgwg mailing list -- [email protected]
To unsubscribe send an email to [email protected]
