Re: [Bloat] bufferbloat at high edge rates

[Stefan Alfredsson <Stefan.Alfredsson@kau.se>]

* Quoting Stefan Alfredsson <stefan.alfredsson@kau.se> [16 Nov-16 07:00]:
> - Firefox tests were run yesterday at around lunchtime  (~12.00 CET), and
> CLI tests just now (~06.40 CET). So time-of-day effect may be a reason for
> less bloat now. I'll do a better comparison when I get to my desktop.

I've done tests with CLI, Firefox and Chrome; repeated 3 times in
a round-robin fashion (cli, ff, chrome, cli, ff, chrome, cli, ff, chrome).

In summary:

- There is 5-10 ms additional latency during all tests.
- CLI and Chrome have the same latency figures, 23 ms base latency 
  Firefox has a base latency of 100 ms RTT, most likely due to US
  responders being used by Firefox and EU responders being used by CLI and Chrome.  
- Chrome achives 30-40 Mbps higher throughput than CLI.


Results from a standard Ubuntu Xenial system running kernel 4.4.0-45-generic #66-Ubuntu SMP Wed Oct 19 14:12:37 UTC 2016

Dslrcli version 0.1 - 15-Nov-2016
test 1 http://www.dslreports.com/speedtest/6169686
test 4 http://www.dslreports.com/speedtest/6169764
test 7 http://www.dslreports.com/speedtest/6169798


Firefox 49.0.2:
test 2 http://www.dslreports.com/speedtest/6169706
test 5 http://www.dslreports.com/speedtest/6169776
test 8 http://www.dslreports.com/speedtest/6169819


Google Chrome 54.0.2840.71
test 3 http://www.dslreports.com/speedtest/6169747
test 6 http://www.dslreports.com/speedtest/6169788
test 9 http://www.dslreports.com/speedtest/6169833



> A better option would be using CAP_NET_RAW, I'll see if this works instead
> of running with full root privs.

Yes, copying the binary to /usr/bin and setting cap_net_raw made it report bloat.
(sudo setcap cap_net_raw+ep /usr/bin/dslrcli-linux-amd64)

/Stefan

-- 
Stefan Alfredsson, PhD         Tel: +46 (0) 54 700 1668
Datavetenskap                  Kontor: 21E-414 (Hus Vanern)
Karlstads universitet          PGP 0xB19B4B16
SE-651 88 Karlstad             http://www.cs.kau.se/stefalfr/