* Re: [Cake] testing cake at 55mbit down/5mbit up
2015-05-05 19:43 [Cake] testing cake at 55mbit down/5mbit up Dave Taht
@ 2015-05-05 19:56 ` Dave Taht
2015-05-05 20:17 ` Dave Taht
2015-05-06 23:48 ` Jonathan Morton
2 siblings, 0 replies; 4+ messages in thread
From: Dave Taht @ 2015-05-05 19:56 UTC (permalink / raw)
To: cake
The behavior is repeatable. I guess I will have to do a detailed exploration
of the captures. Perhaps this kernel or the one on the remote server has issues.
Linux lakshmi 3.16.0-31-lowlatency #43-Ubuntu SMP PREEMPT Tue Mar 10
17:56:21 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux
On Tue, May 5, 2015 at 12:43 PM, Dave Taht <dave.taht@gmail.com> wrote:
> http://snapon.lab.bufferbloat.net/~d/prio_cake/
>
> I did have one massively anomalous result, on the 6vs4compete test,
> where it had a crazy high number of drops.
>
> The rest looked pretty sane although the diffserv4 thing did not line
> up with the markings netperf-wrapper is using for the rrul test.
>
> qdisc cake 801f: root refcnt 2 bandwidth 5Mbit diffserv4 flows
> Sent 40103696 bytes 150518 pkt (dropped 11579, overlimits 449993 requeues 0)
> backlog 0b 0p requeues 0
> Class 0 Class 1 Class 2 Class 3
> rate 5Mbit 4687Kbit 3750Kbit 1250Kbit
> target 5.0ms 5.0ms 5.0ms 14.5ms
> interval 105.0ms 105.0ms 105.0ms 116.3ms
> Pk delay 52.7ms 7.5ms 1.9ms 1.1ms
> Av delay 21.5ms 4.5ms 118us 39us
> Sp delay 2.3ms 1us 108us 20us
> pkts 66383 95449 103 162
> way inds 0 0 0 0
> way miss 5 39 103 5
> way cols 0 0 0 0
> bytes 9484277 31571621 11481 10311
> drops 11572 7 0 0
> marks 2429 745 0 0
> qdisc ingress ffff: parent ffff:fff1 ----------------
> Sent 406045781 bytes 293044 pkt (dropped 0, overlimits 0 requeues 0)
> backlog 0b 0p requeues 0
>
>
> --
> Dave Täht
> Open Networking needs **Open Source Hardware**
>
> https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
--
Dave Täht
Open Networking needs **Open Source Hardware**
https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Cake] testing cake at 55mbit down/5mbit up
2015-05-05 19:43 [Cake] testing cake at 55mbit down/5mbit up Dave Taht
2015-05-05 19:56 ` Dave Taht
@ 2015-05-05 20:17 ` Dave Taht
2015-05-06 23:48 ` Jonathan Morton
2 siblings, 0 replies; 4+ messages in thread
From: Dave Taht @ 2015-05-05 20:17 UTC (permalink / raw)
To: cake
and it is also repeatable from an osx box, over wifi.
root@cake:~# tc -s qdisc show dev eth0
qdisc cake 803d: root refcnt 2 bandwidth 5Mbit diffserv4 flows
Sent 4794358 bytes 11957 pkt (dropped 32, overlimits 28003 requeues 0)
backlog 99395b 73p requeues 0
Class 0 Class 1 Class 2 Class 3
rate 5Mbit 4687Kbit 3750Kbit 1250Kbit
target 5.0ms 5.0ms 5.0ms 14.5ms
interval 105.0ms 105.0ms 105.0ms 116.3ms
Pk delay 323.6ms 153.2ms 2us 2.7ms
Av delay 59.1ms 52.0ms 0us 883us
Sp delay 6.1ms 3.3ms 0us 152us
pkts 4701 6658 9 694
way inds 0 0 0 0
way miss 5 40 9 2
way cols 0 0 0 0
bytes 1831961 3060910 999 43749
drops 22 10 0 0
marks 270 231 0 0
qdisc ingress ffff: parent ffff:fff1 ----------------
Sent 50484063 bytes 38474 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
On Tue, May 5, 2015 at 12:43 PM, Dave Taht <dave.taht@gmail.com> wrote:
> http://snapon.lab.bufferbloat.net/~d/prio_cake/
>
> I did have one massively anomalous result, on the 6vs4compete test,
> where it had a crazy high number of drops.
>
> The rest looked pretty sane although the diffserv4 thing did not line
> up with the markings netperf-wrapper is using for the rrul test.
>
> qdisc cake 801f: root refcnt 2 bandwidth 5Mbit diffserv4 flows
> Sent 40103696 bytes 150518 pkt (dropped 11579, overlimits 449993 requeues 0)
> backlog 0b 0p requeues 0
> Class 0 Class 1 Class 2 Class 3
> rate 5Mbit 4687Kbit 3750Kbit 1250Kbit
> target 5.0ms 5.0ms 5.0ms 14.5ms
> interval 105.0ms 105.0ms 105.0ms 116.3ms
> Pk delay 52.7ms 7.5ms 1.9ms 1.1ms
> Av delay 21.5ms 4.5ms 118us 39us
> Sp delay 2.3ms 1us 108us 20us
> pkts 66383 95449 103 162
> way inds 0 0 0 0
> way miss 5 39 103 5
> way cols 0 0 0 0
> bytes 9484277 31571621 11481 10311
> drops 11572 7 0 0
> marks 2429 745 0 0
> qdisc ingress ffff: parent ffff:fff1 ----------------
> Sent 406045781 bytes 293044 pkt (dropped 0, overlimits 0 requeues 0)
> backlog 0b 0p requeues 0
>
>
> --
> Dave Täht
> Open Networking needs **Open Source Hardware**
>
> https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
--
Dave Täht
Open Networking needs **Open Source Hardware**
https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [Cake] testing cake at 55mbit down/5mbit up
2015-05-05 19:43 [Cake] testing cake at 55mbit down/5mbit up Dave Taht
2015-05-05 19:56 ` Dave Taht
2015-05-05 20:17 ` Dave Taht
@ 2015-05-06 23:48 ` Jonathan Morton
2 siblings, 0 replies; 4+ messages in thread
From: Jonathan Morton @ 2015-05-06 23:48 UTC (permalink / raw)
To: Dave Taht; +Cc: cake
> On 5 May, 2015, at 22:43, Dave Taht <dave.taht@gmail.com> wrote:
>
> the diffserv4 thing did not line
> up with the markings netperf-wrapper is using for the rrul test.
True, but neither does Wi-Fi QoS, which ignores the low-order DSCP bits and uses only the Precedence ones, with some re-ordering:
Precedence 0 & 3 go into BE,
1 & 2 into BK,
4 & 5 into VI,
6 & 7 into VO.
Both CS5 and EF therefore fall into the VI (video) queue on Wi-Fi; only CS6 and above get the VO (voice) queue. This doesn’t make much sense, which is probably one reason why I ignored that practice when setting up cake.
With cake, CS5 and EF both fall into the "voice" class; CS5 supposedly corresponds to “voice signalling”, which is why I put it there. Cake’s “video” class is meant for applications which need reliable access to bandwidth (eg. streaming video) rather than especially low latency.
Here’s a summary of where all the well-known DSCPs end up on Wi-Fi and cake:
DSCP WiFi Cake
==== ==== ====
CS0 BE BE <- default
TOS1 BE VI <- max reliability (legacy)
TOS2 BE BE <- max throughput (legacy)
TOS4 BE VI <- min delay (legacy)
CS1 BK BK <- low priority
AF1x BK BE <- high throughput
CS2 BK VI <- admin & management
AF2x BK VI <- low latency data, ie. database
CS3 BE VI <- broadcast video
AF3x BE VI <- multimedia streaming, ie. YouTube
CS4 VI VO <- realtime interactive, ie. games
AF4x VI VI <- multimedia conferencing
CS5 VI VO <- signalling
VA VI VO <- telephony with capacity reservation
EF VI VO <- VoIP and games
CS6 VO VO <- network control, ie. NTP
CS7 VO VO <- network control, ie. routing
- Jonathan Morton
^ permalink raw reply [flat|nested] 4+ messages in thread