justin: thx for nuking the log scale. that makes the bloat much more visible here (typical cablemodem)
http://www.dslreports.com/speedtest/322800 I am puzzled as to my post fq_codel result here at T+40 and will have to repeat... http://www.dslreports.com/speedtest/322992 On Sun, Apr 19, 2015 at 12:22 PM, Dave Taht <[email protected]> wrote: > So my "acid test", taken on a linux box over ethernet, through a > cerowrt box with sqm-scripts turned off. This is my gf's comcast > "blast" service, which is rated for 55mbits down and 5.5mbits up. The > new speedtest does indeed show results that have the typical bloat > (nearly a second) a cable modem has when tested solely for up and > down, separately. > > http://www.dslreports.com/speedtest/322800 > > A) I definitely am not particularly huge on defaulting to a log scale > for this graph. Or rather, I would be huge on the graph defaulting to > a linear scale AND huge when you get numbers as bad as this. :) > > B) is there a way to specify ipv6? > > 2) So I did a follow on speedtest with fq_codel enabled to shape with > sqm. it may be that my upload shaper is a bit over what is desirable > for this link, and I need to repeat. > > http://www.dslreports.com/speedtest/322992 > > 3) For giggles, this one is with ecn enabled both on the shaper and on > the tcp I am using, showing that this speedtest site will use ecn when > enabled: > > http://www.dslreports.com/speedtest/323101 > > Some ecn mark results on the shaper for that: > > http://pastebin.com/sniePC1M > > Both these tests are showing some latency spikes so it does look like > I should tune down the shaper a bit. > > 3) The rrul, rrul_be, tcp_upload, and tcp_download test data from > netperf-wrapper under both these circumstances (no ecn) is up at: > > http://snapon.lab.bufferbloat.net/~d/lorna-ethernet.tgz > > Feel free to create graphs and comparisons to suit selves. > > A few graphs: > cdf plot of latency (have to use a log scale!!) compared between the > shaped and unshaped alternatives... > > http://snapon.lab.bufferbloat.net/~d/lorna-ethernet/comcast_55_speedtest_comparison.png > > what the overbuffered download looks like: > > http://snapon.lab.bufferbloat.net/~d/lorna-ethernet/comcast_55_speedtest_comparison_download_bloated.png > > The shaped fq_codeled equivalent: > http://snapon.lab.bufferbloat.net/~d/lorna-ethernet/comcast_55_speedtest_comparison_download_fq_codeled.png > > (log scale comparison again - 10ms of induced latency vs 400+! Aggh) > > And a graph of probably minimal usefulness comparing the behavior of > the shaped vs unshaped download flow configurations, for both ipv4 and > ipv6. > > http://snapon.lab.bufferbloat.net/~d/lorna-ethernet/comcast_55_speedtest_comparison_download_compared.png > > On Sun, Apr 19, 2015 at 10:43 AM, Dave Taht <[email protected]> wrote: >> I am going to do an acid test today. The line I tested last night is a >> comcast cable line (with htb+fq_codel on the link). So I plan to plug >> in the ethernet on both my mac and linux laptops and repeat the >> comparison with the shaper on and off, with both linux and osx. >> >> the *really funny* part of this is that I do not have a single extra >> ethernet cable in my gf's SF apartment, and the less funny part of >> this is the nearest radio shack is now closed.... > > > > -- > 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 _______________________________________________ Bloat mailing list [email protected] https://lists.bufferbloat.net/listinfo/bloat
