[Bloat] [Starlink] an rtt_fair test?
Dave Taht
dave.taht at gmail.com
Sun Apr 17 20:22:02 EDT 2022
On Sun, Apr 17, 2022 at 8:36 AM <jordan at inacomptc.com> wrote:
>
> You're welcome! Here's the files Flent generated: https://1drv.ms/u/s!Ap4u4Rte63FqjqoSQXX444OWLvl8FQ?e=f22Nt7
Thank you for the data!
Part of with me unleashing this test on y'all is to validate my cloud,
which was looking weird yesterday. I still haven't quite got to the
problem fremont or london actually had, but it's much better now.
> And thanks for the link! It honestly doesn't look too bad to implement so I might give it a shot today if time allows.
Honestly, I hope y'all are out there spending time with your families
today. The Net can wait.
>
> I've also been contemplating replacing it with something like a Mikrotik since cake is now integrated into them. I've been testing one at home and made a post with a bunch of flent tests to the cake thread in their forum if you're interested in some data from a Charter Spectrum cable connection: https://forum.mikrotik.com/viewtopic.php?t=179307#p925485
That is a really wonderful addition to that thread, thank you. For
those here that haven't gone through it, the pictures help a lot, but
requires you register for a free account. Perhaps at some point we can
eliminate the confusion and noise from it, but I often find it useful
to deeply understand what mistakes others will make. ( I have no idea
why anyone is using the bucket thing on that bug thread... no buckets
needed. I don't get it)
Seeing that horrible 1sec of induced latency on spectrum and knowing
that they have 28m subscribers whose networks still behave like that,
pains me. I would so hope that their gbit service, at least, was only
in the 120ms range.
Pro tip: tc -s qdisc show
shows you the drop/market/reschedules
on the client, you might see a few reschedules under the structure of
this test but no drops as "TCP small queues" keeps that together. On
router, the drop ratios are geometrically relative to the bandwidth.
If I have any one gripe left about the mikrotik release it is that
they make backlog, packet and ,drop statistics easily available across
their OSes. I'm pretty sure that at higher bandwidths their default
settings for SFQ and pfifo are too low...
>
> The OneDrive link to my flent files should still be active in that post as well.
>
> -Jordan
>
> -----Original Message-----
> From: Dave Taht <dave.taht at gmail.com>
> Sent: Sunday, April 17, 2022 10:57 AM
> To: Jordan Szuch <jordan at inacomptc.com>
> Cc: flent-users <flent-users at flent.org>; bloat <bloat at lists.bufferbloat.net>; starlink at lists.bufferbloat.net
> Subject: Re: [Starlink] an rtt_fair test?
>
> Thank you for testing!!! If you could stash the *.flent.gz files somewhere I could get at, that would be great. You can also browse the plots using the flent-gui to see more detail over time.
>
> Cake, long available as an out of tree build for the edgerouter, has both dsl compensation and an ack-filter which would help A LOT, in your case. i wish I'd pushed harder to get dsl compensation into the edgerouter smart queues code...
>
> https://community.ui.com/questions/Cake-compiled-for-the-EdgeRouter-devices/fc1ff27c-f321-4344-8737-fcc755cae8a2
>
> Given your 500kbit upload speed the minimum latency inflation under the structure of this test (4 upload flows) should have been around 150ms (4 * 25ms + ack overhead), but without looking at the *.gz files I don't know what your baseline latency was to any of these sites. (it would be good if flent reported on the text output what the minimum actually was)
>
> On Sun, Apr 17, 2022 at 7:35 AM <jordan at inacomptc.com> wrote:
> >
> > Hi Dave,
> >
> > This isn't specifically cake related but I figured it could be interesting, especially as these are from a location where I've been investigating setting up something with cake. These are statistics from a relatives location where they only have a DSL line from CenturyLink available to them. There's a little Debian based VM I setup for a couple random things so I was able to run flent from that. They have a EdgeRouter X serving as the router so they can at least take advantage of fq_codel through the routers Smart Queue feature. I ran a quick speed test, then ran flent without sqm, and then with sqm. Sorry if this gets a bit long and hopefully it stays readable.
> >
> > #############
> > Speedtest by Ookla
> >
> > Server: 123.Net, Inc. - Southfield, MI (id = 15044)
> > ISP: CenturyLink
> > Latency: 45.45 ms (194.34 ms jitter)
> > Download: 8.88 Mbps (data used: 12.6 MB )
> > Upload: 0.64 Mbps (data used: 962.7 kB )
> > Packet Loss: 5.1%
> > Result URL:
> > https://www.speedtest.net/result/c/799ba46b-d3d4-4d54-ac76-118fac847b7
> > 3
> >
> > ###########
> > No SQM or QOS
> >
> > flent -x --socket-stats --step-size=.05 -H de.starlink.taht.net -H
> > london.starlink.taht.net -H singapore.starlink.taht.net -H fremont.starlink.taht.net -t nosqm_erx_rttfair4be_8Mbpsdown_500Kbpsup_centurylinkdsl rtt_fair4be Started Flent 2.0.0 using Python 3.9.2.
> > Starting rtt_fair4be test. Expected run time: 70 seconds.
> > Data file written to
> > ./rtt_fair4be-2022-04-17T145553.865990.nosqm_erx_rttfair4be_8Mbpsdown_
> > 500Kbpsup_centurylinkdsl.flent.gz
> >
> > Summary of rtt_fair4be test run from 2022-04-17 13:55:53.865990
> > Title: 'nosqm_erx_rttfair4be_8Mbpsdown_500Kbpsup_centurylinkdsl'
> >
> > avg median # data pts
> > Ping (ms) ICMP1 de.starlink.taht.net : 1026.06 920.00 ms 1383
> > Ping (ms) ICMP2 london.starlink.taht.net : 1009.28 887.00 ms 1383
> > Ping (ms) ICMP3 singapore.starlink.taht.net : 1121.09 1012.00 ms 1383
> > Ping (ms) ICMP4 fremont.starlink.taht.net : 943.77 853.00 ms 1383
> > Ping (ms) avg : 1025.05 N/A ms 1383
> > TCP download BE1 de.starlink.taht.net : 1.36 2.00 Mbits/s 1383
> > TCP download BE2 london.starlink.taht.net : 0.72 0.65 Mbits/s 1383
> > TCP download BE3 singapore.starlink.taht.net : 1.03 1.74 Mbits/s 1383
> > TCP download BE4 fremont.starlink.taht.net : 2.11 2.88 Mbits/s 1383
> > TCP download avg : 1.30 N/A Mbits/s 1383
> > TCP download fairness : 0.86 N/A Mbits/s 1383
> > TCP download sum : 5.22 N/A Mbits/s 1383
> > TCP upload BE1 de.starlink.taht.net : 0.12 0.20 Mbits/s 1383
> > TCP upload BE1 de.starlink.taht.net::tcp_cwnd : 16.85 14.00 1044
> > TCP upload BE1 de.starlink.taht.net::tcp_delivery_rate : 0.12 0.12 1036
> > TCP upload BE1 de.starlink.taht.net::tcp_pacing_rate : 0.35 0.26 1042
> > TCP upload BE1 de.starlink.taht.net::tcp_rtt : 1128.57 954.47 1020
> > TCP upload BE1 de.starlink.taht.net::tcp_rtt_var : 68.22 50.08 1020
> > TCP upload BE2 london.starlink.taht.net : 0.12 0.15 Mbits/s 1383
> > TCP upload BE2 london.starlink.taht.net::tcp_cwnd : 17.33 11.00 1046
> > TCP upload BE2 london.starlink.taht.net::tcp_delivery_rate : 0.12 0.11 1041
> > TCP upload BE2 london.starlink.taht.net::tcp_pacing_rate : 0.38 0.26 1044
> > TCP upload BE2 london.starlink.taht.net::tcp_rtt : 1166.74 991.48 1043
> > TCP upload BE2 london.starlink.taht.net::tcp_rtt_var : 79.99 68.27 1043
> > TCP upload BE3 singapore.starlink.taht.net : 0.07 0.13 Mbits/s 1383
> > TCP upload BE3 singapore.starlink.taht.net::tcp_cwnd : 13.60 9.00 1042
> > TCP upload BE3 singapore.starlink.taht.net::tcp_delivery_rate : 0.07 0.08 1013
> > TCP upload BE3 singapore.starlink.taht.net::tcp_pacing_rate : 0.23 0.20 1031
> > TCP upload BE3 singapore.starlink.taht.net::tcp_rtt : 1302.23 1129.07 967
> > TCP upload BE3 singapore.starlink.taht.net::tcp_rtt_var : 113.67 98.84 967
> > TCP upload BE4 fremont.starlink.taht.net : 0.11 0.20 Mbits/s 1383
> > TCP upload BE4 fremont.starlink.taht.net::tcp_cwnd : 17.55 11.00 1038
> > TCP upload BE4 fremont.starlink.taht.net::tcp_delivery_rate : 0.11 0.10 1035
> > TCP upload BE4 fremont.starlink.taht.net::tcp_pacing_rate : 0.32 0.23 1036
> > TCP upload BE4 fremont.starlink.taht.net::tcp_rtt : 1094.64 887.54 1033
> > TCP upload BE4 fremont.starlink.taht.net::tcp_rtt_var : 67.20 59.16 1033
> > TCP upload avg : 0.10 N/A Mbits/s 1383
> > TCP upload fairness : 0.96 N/A Mbits/s 1383
> > TCP upload sum : 0.42 N/A Mbits/s 1383
> >
> > ##########
> > with smart queue (fq_codel) on the EdgeRouter
> >
> > flent -x --socket-stats --step-size=.05 -H de.starlink.taht.net -H
> > london.starlink.taht.net -H singapore.starlink.taht.net -H fremont.starlink.taht.net -t fqcodel_erx_rttfair4be_8Mbpsdown_500Kbpsup_centurylinkdsl rtt_fair4be Started Flent 2.0.0 using Python 3.9.2.
> > Starting rtt_fair4be test. Expected run time: 70 seconds.
> > Data file written to
> > ./rtt_fair4be-2022-04-17T150339.549817.fqcodel_erx_rttfair4be_8Mbpsdow
> > n_500Kbpsup_centurylinkdsl.flent.gz
> >
> > Summary of rtt_fair4be test run from 2022-04-17 14:03:39.549817
> > Title: 'fqcodel_erx_rttfair4be_8Mbpsdown_500Kbpsup_centurylinkdsl'
> >
> > avg median # data pts
> > Ping (ms) ICMP1 de.starlink.taht.net : 392.93 208.00 ms 1384
> > Ping (ms) ICMP2 london.starlink.taht.net : 373.94 196.00 ms 1384
> > Ping (ms) ICMP3 singapore.starlink.taht.net : 508.15 319.00 ms 1384
> > Ping (ms) ICMP4 fremont.starlink.taht.net : 346.23 152.00 ms 1384
> > Ping (ms) avg : 405.31 N/A ms 1384
> > TCP download BE1 de.starlink.taht.net : 1.36 1.62 Mbits/s 1384
> > TCP download BE2 london.starlink.taht.net : 1.47 1.63 Mbits/s 1384
> > TCP download BE3 singapore.starlink.taht.net : 1.55 1.60 Mbits/s 1384
> > TCP download BE4 fremont.starlink.taht.net : 1.50 1.80 Mbits/s 1384
> > TCP download avg : 1.47 N/A Mbits/s 1384
> > TCP download fairness : 1.00 N/A Mbits/s 1384
> > TCP download sum : 5.88 N/A Mbits/s 1384
> > TCP upload BE1 de.starlink.taht.net : 0.09 0.85 Mbits/s 1384
> > TCP upload BE1 de.starlink.taht.net::tcp_cwnd : 4.50 4.00 1035
> > TCP upload BE1 de.starlink.taht.net::tcp_delivery_rate : 0.08 0.08 1030
> > TCP upload BE1 de.starlink.taht.net::tcp_pacing_rate : 0.17 0.15 1032
> > TCP upload BE1 de.starlink.taht.net::tcp_rtt : 458.11 318.82 1027
> > TCP upload BE1 de.starlink.taht.net::tcp_rtt_var : 95.76 61.88 1027
> > TCP upload BE2 london.starlink.taht.net : 0.09 0.93 Mbits/s 1384
> > TCP upload BE2 london.starlink.taht.net::tcp_cwnd : 4.17 4.00 1038
> > TCP upload BE2 london.starlink.taht.net::tcp_delivery_rate : 0.08 0.08 1033
> > TCP upload BE2 london.starlink.taht.net::tcp_pacing_rate : 0.17 0.14 1036
> > TCP upload BE2 london.starlink.taht.net::tcp_rtt : 452.67 322.49 1033
> > TCP upload BE2 london.starlink.taht.net::tcp_rtt_var : 96.87 64.28 1033
> > TCP upload BE3 singapore.starlink.taht.net : 0.06 0.44 Mbits/s 1384
> > TCP upload BE3 singapore.starlink.taht.net::tcp_cwnd : 4.16 4.00 1040
> > TCP upload BE3 singapore.starlink.taht.net::tcp_delivery_rate : 0.06 0.06 1031
> > TCP upload BE3 singapore.starlink.taht.net::tcp_pacing_rate : 0.12 0.11 1036
> > TCP upload BE3 singapore.starlink.taht.net::tcp_rtt : 577.75 438.61 1027
> > TCP upload BE3 singapore.starlink.taht.net::tcp_rtt_var : 107.05 72.97 1027
> > TCP upload BE4 fremont.starlink.taht.net : 0.10 1.48 Mbits/s 1384
> > TCP upload BE4 fremont.starlink.taht.net::tcp_cwnd : 4.45 4.00 1020
> > TCP upload BE4 fremont.starlink.taht.net::tcp_delivery_rate : 0.09 0.09 1017
> > TCP upload BE4 fremont.starlink.taht.net::tcp_pacing_rate : 0.21 0.16 1019
> > TCP upload BE4 fremont.starlink.taht.net::tcp_rtt : 439.05 285.86 1018
> > TCP upload BE4 fremont.starlink.taht.net::tcp_rtt_var : 81.90 56.33 1018
> > TCP upload avg : 0.08 N/A Mbits/s 1384
> > TCP upload fairness : 0.97 N/A Mbits/s 1384
> > TCP upload sum : 0.34 N/A Mbits/s 1384
> >
> >
> > EdgeRouter Settings
> >
> > smart-queue DSL {
> > download {
> > ecn enable
> > flows 1024
> > fq-quantum 300
> > limit 10240
> > rate 8mbit
> > }
> > upload {
> > ecn enable
> > flows 1024
> > fq-quantum 300
> > limit 10240
> > rate 600kbit
> > target 15ms
> > }
> > wan-interface pppoe0
> >
> > Also, here's the VMs qdisc settings. Which I now realize probably
> > aren't the best
> >
> > sysctl -a | grep qdisc
> > net.core.default_qdisc = pfifo_fast
> >
> > tc -s qdisc
> > qdisc noqueue 0: dev lo root refcnt 2
> > Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0) backlog 0b
> > 0p requeues 0 qdisc mq 0: dev eth0 root Sent 628201 bytes 8643 pkt
> > (dropped 0, overlimits 0 requeues 0) backlog 0b 0p requeues 0 qdisc
> > pfifo_fast 0: dev eth0 parent :1 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1
> > 1 1 1 1 1 Sent 628201 bytes 8643 pkt (dropped 0, overlimits 0
> > requeues 0) backlog 0b 0p requeues 0
> >
> >
> >
> > -Jordan
> >
> > -----Original Message-----
> > From: Starlink <starlink-bounces at lists.bufferbloat.net> On Behalf Of
> > Dave Taht
> > Sent: Saturday, April 16, 2022 9:40 PM
> > To: flent-users <flent-users at flent.org>; bloat
> > <bloat at lists.bufferbloat.net>; starlink at lists.bufferbloat.net
> > Subject: [Starlink] an rtt_fair test?
> >
> > if anyone has a bit of spare time this holiday, I'd like to collect a few results from various networks around the web this week, not just on the turris. Example command line and outputs here:
> >
> > https://forum.turris.cz/t/sqm-on-turris-flent-benchmarks/17048/
> >
> > --
> > I tried to build a better future, a few times:
> > https://wayforward.archive.org/?site=https%3A%2F%2Fwww.icei.org
> >
> > Dave Täht CEO, TekLibre, LLC
> > _______________________________________________
> > Starlink mailing list
> > Starlink at lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo/starlink
> >
>
>
> --
> I tried to build a better future, a few times:
> https://wayforward.archive.org/?site=https%3A%2F%2Fwww.icei.org
>
> Dave Täht CEO, TekLibre, LLC
>
--
I tried to build a better future, a few times:
https://wayforward.archive.org/?site=https%3A%2F%2Fwww.icei.org
Dave Täht CEO, TekLibre, LLC
More information about the Bloat
mailing list