Thread (6 messages) flat view 6 messages, 5 authors, 2021-09-26

Re: [PATCH net-next] tcp: tracking packets with CE marks in BW rate sample

From: Dave Taht <hidden>
Date: 2021-09-26 22:33:21

On Fri, Sep 24, 2021 at 4:43 PM Yuchung Cheng [off-list ref] wrote:
On Fri, Sep 24, 2021 at 1:20 PM Jakub Kicinski [off-list ref] wrote:
quoted
On Thu, 23 Sep 2021 21:17:07 +0000 Luke Hsiao wrote:
quoted
From: Yuchung Cheng <redacted>

In order to track CE marks per rate sample (one round trip), TCP needs a
per-skb header field to record the tp->delivered_ce count when the skb
was sent. To make space, we replace the "last_in_flight" field which is
used exclusively for NV congestion control. The stat needed by NV can be
alternatively approximated by existing stats tcp_sock delivered and
mss_cache.

This patch counts the number of packets delivered which have CE marks in
the rate sample, using similar approach of delivery accounting.
Is this expected to be used from BPF CC? I don't see a user..
Great question. Yes the commit message could be more clear that this
intends for both ebpf-CC or other third party module that use ECN. For
example bbr2 uses it heavily (bbr2 upstream WIP). This feature is
useful for congestion control research which many use ECN as core
signals now.
I am glad a common API to this is emerging. RFC3168 compliant aqms
(fq_codel, pie, cake) can also emit multiple CE marks per RTT, and
the defined response to them is inadequate. Glad you found space!

There is also this ecn problem outstanding elsewhere, that I hope
bbrv2 has looked into?

https://www.bobbriscoe.net/projects/latency/sub-mss-w.pdf

Otherwise:

Reviewed-by: Dave Taht <redacted>


--
Fixing Starlink's Latencies: https://www.youtube.com/watch?v=c9gLo6Xrwgw

Dave Täht CEO, TekLibre, LLC
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help