net_sched: sch_fq: add dctcp-like marking
authorEric Dumazet <edumazet@google.com>
Sun, 11 Nov 2018 17:11:31 +0000 (09:11 -0800)
committerDavid S. Miller <davem@davemloft.net>
Sun, 11 Nov 2018 21:59:21 +0000 (13:59 -0800)
commit48872c11b77271ef9b070bdc50afe6655c4eb9aa
treecebc970abf4ab61ae117ef21b18a3b967eb9a481
parentc73e5807e4f6fc6d373a5db55b45f639f8bb6328
net_sched: sch_fq: add dctcp-like marking

Similar to 80ba92fa1a92 ("codel: add ce_threshold attribute")

After EDT adoption, it became easier to implement DCTCP-like CE marking.

In many cases, queues are not building in the network fabric but on
the hosts themselves.

If packets leaving fq missed their Earliest Departure Time by XXX usec,
we mark them with ECN CE. This gives a feedback (after one RTT) to
the sender to slow down and find better operating mode.

Example :

tc qd replace dev eth0 root fq ce_threshold 2.5ms

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/uapi/linux/pkt_sched.h
net/sched/sch_fq.c