Hello list,
I'm on AA 12.09 using qos-scripts 1.2.1-6. The config is pretty much
default, see below.
I want video streams (http streams, e.g. newscasts and trailers) to load
quickly and play smoothly, even when there's loads of bulk traffic, e.g.
multiple ftp transfers.
So I let HTTP traffic default to the "normal" bucket. And then I put all
the traffic that can wait into the "bulk" bucket.
The last part is what only partly works. For egress traffic it works fine. But
ingress traffic never visits the bulk bucket (1:40). I checked with all sorts
of traffic, e.g. rsync, pop3s, smtps and ftp:
# Interface: wan
# Direction: Egress
# Stats: Start
...
class hfsc 1:30 parent 1:1 leaf 300: ls m1 0bit d 100.0ms m2 90000bit ul m1
0bit d 0us m2 650000bit
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
period 0 level 0
class hfsc 1:40 parent 1:1 leaf 400: ls m1 0bit d 200.0ms m2 18000bit ul m1
0bit d 0us m2 650000bit
Sent 1119192 bytes 19956 pkt (dropped 0, overlimits 0 requeues 0) <--- on
egress it works
backlog 0b 0p requeues 0
period 15983 work 1119192 bytes level 0
# Interface: wan
# Direction: Ingress
# Stats: Start
...
class hfsc 1:30 parent 1:1 leaf 300: ls m1 0bit d 100.0ms m2 1180Kbit ul m1
0bit d 0us m2 8500Kbit
Sent 50491942 bytes 33440 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
period 6473 work 50101816 bytes level 0
class hfsc 1:40 parent 1:1 leaf 400: ls m1 0bit d 200.0ms m2 236000bit ul m1
0bit d 0us m2 8500Kbit
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0) <--- on ingress: FAIL
backlog 0b 0p requeues 0
period 0 level 0
Let's say I have multiple ftp downloads ongoing at full speed and then I want
to watch an HD trailer. I would expect that the ftp traffic ends up in the
"bulk" bucket. http gets put into the "default" bucket, which has a higher
priority. Then the queuing discipline starts dropping ftp packets and the ftp
server slows down. The http stream gets the attention I want it to get.
Unfortunately I doesn't yet work like that. The initial buffering of the http
stream takes ages. And once it starts up it's prone to pause again and buffer
some more. Currently the only way for me to enjoy a stream is to dial down the
bulk transfers manually.
Has anybody an idea what's wrong?
Bonus question: Why does the "mark" value in /proc/net/nf_conntrack
appear to be unreliable?
Example: I ping some site. ICMP traffic should go to the priority bucket, so I
expect it to get marked "1/0xff", yet the mark value is "0".
Here is my configuration (I kept the original classgroup and classes setup as
they were):
# INTERFACES:
config interface wan
option classgroup "Default"
option enabled 1
option upload 650
option download 8500
# RULES:
config classify
option target "Priority"
option ports "22,53"
config default
option target "Express"
option proto "udp"
option pktsize "-500"
config reclassify
option target "Priority"
option proto "icmp"
config default
option target "Bulk"
option ports "20,21,465,873,995"
option proto "tcp"
config default
option target "Bulk"
option portrange "1024-65535"
config reclassify
option target "Priority"
option proto "tcp"
option pktsize "-128"
option mark "!Bulk"
option tcpflags "SYN"
config reclassify
option target "Priority"
option proto "tcp"
option pktsize "-128"
option mark "!Bulk"
option tcpflags "ACK"
Thanks for reading!
Kind regards,
Sebastian
_______________________________________________
openwrt-users mailing list
[email protected]
https://lists.openwrt.org/cgi-bin/mailman/listinfo/openwrt-users
