Hi,
I noticed while debugging a VPN QoS performance issue that my STAB
overheads are not getting applied correctly.
I have tc stab configured as:
tc qdisc add dev eno1 root stab overhead 30 mpu 68 handle 1:0 estimator
1sec 4sec hfsc default 14
1444 byte packets originating from an ipsec VPN are exiting via eno1.
They are classified into 1:11 via iptables fwmark which is caught by tc
filter. There is other small traffic in 1:11 such as tcp acks, small
udp, etc but the largest packets have been confirmed to be the ones
originating via the VPN.
I've confirmed the size of the packets arriving at the destination
server and they are 1444 bytes therefore tc should be counting these as
1444+30. However it isnt:
qdisc fq_codel 1011: parent 1:11 limit 10240p flows 1024 quantum 1514
target 5.0ms interval 50.0ms memory_limit 32Mb
Sent 57965977444 bytes 44285275 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
maxpacket 1444 drop_overlimit 0 new_flow_count 1214436 ecn_mark 0
new_flows_len 0 old_flows_len 1
Box is a CentOS 7 server with Plus kernel
3.10.0-957.27.2.el7.centos.plus.bug16242.x86_64.
Alan
