[Cerowrt-devel] [Cake] some reference numbers for cake performance on x86_64

[Sebastian Moeller]

Hi Dave, hi list,


On Jun 29, 2015, at 01:42 , Dave Taht <dave.taht at gmail.com> wrote:

> This is a set of two hop (client-server through switch) results, with
> sch_fq on the nuc client, and cake on the rangeley (ranger) server.
> 
> Since we seem to be aiming for sane results from the mvneta, at speeds
> at 500mbit...
> 
> http://snapon.lab.bufferbloat.net/~d/k500/
> 
> Do note that this is a live system with other traffic so it is noisy data.
> 
> Of note, in looking at this dataset in comparison to mikaels, is that
> he only got 360mbit bidirectional on rrul_be - where I get full
> throughput. admittedly he seemed to be running using htb as the rate
> limiter. Being off by *20%* from the set setting (after subtracting
> acks) of 500 mbit however, seems to be an issue.

So ignoring the ACKs we can expect a TCP/IPv4 payload rate up to 480 Mbps.

TCP/IPv4 Payload at 500Mbps shaping:
1500 - 20 - 20 = 1460 Byte
payload to OTWS ratio
1460/1518 = 0.961791831357
Payload Bandwidth
500*1460/1518 = 480.90 Mbps
Packet rate:
500*1000^2 / (1518 * 8) = 41K


In other words a loss of roughly 4 percent for the required protocol headers. For the ACK traffic an rough estimate would be:
one ACK for every other packet (allowing delayed ACKs), increasing the packet rates to roughly 60K per direction and eating:
(20000 * 64 * 8) / 1000^2 = 10.24 Mbps

So a rrul test should give us something close to 470Mbps (ignoring the UDP and ICMP flows and) or roughly 100-100*470/500 = 6% loss not 20%.

On the positive side I seem to remember we were looking for a rower that can handle 300Mbps bi-directionally and it seems the 1200ac cmes floes or even exceeds this threshold ;)

Best Regards
	Sebastian




> 
> I will setup another box to do client - router -server this week.
> 
> -- 
> Dave Täht
> worldwide bufferbloat report:
> http://www.dslreports.com/speedtest/results/bufferbloat
> And:
> What will it take to vastly improve wifi for everyone?
> https://plus.google.com/u/0/explore/makewififast
> _______________________________________________
> Cake mailing list
> Cake at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cake