Re: [tsvwg] Linux DCTCP response to loss (was: L4S and the RACK requirement)

[Bob Briscoe <[e-mail]>]

Yuchung, [also adding Olivier who was about to try to repost Koen's 
patch to Linux netdev]


I'm saying the Linux implementation of DCTCP RFC has a bug (not a design 
issue).
[So this fork of the thread probably ought to move to a more appropriate 
list, given it's Linux-specific.]

I wasn't aware of any problem with Koen's patch reducing by 75% instead 
of 50%.
https://www.spinics.net/lists/netdev/msg420916.html
Are you saying you've tried the patch and it reduces by 75%, or you've 
looked at the code and you think it will reduce by 75%?


Bob


On 22/02/2019 18:15, Yuchung Cheng wrote:
>> For instance "MUST react to packet loss in a way that will coexist safely with a TCP Reno" is a safety requirement. So a DCTCP that does not satisfy this requirement cannot safely be used on the Internet For instance, Windows DCTCP is safe on the public Internet, but current mainline Linux DCTCP is not because there's bug where it intends to fall back to Reno - here's Koen's patch from 2yrs ago.
> how so? I believe Linux DCTCP correctly implements
> https://tools.ietf.org/html/rfc8257#section-3.5
>
> AFAIK Koen's patch makes DCTCP reduces cwnd by 75% (instead of 50%)
> when a loss happens w/o any prior CE marks (in a round trip).
>
> Could you identify which case you're referring to
> 1) DCTCP RFC has a design issue
> 2) Linux implementation of DCTCP RFC has a bug
>
>

-- 
________________________________________________________________
Bob Briscoe                               http://bobbriscoe.net/