Re: [Cake] cake flenter results round 4
On Mon, Dec 4, 2017 at 1:02 AM, Pete Heistwrote: > > On Dec 4, 2017, at 4:29 AM, Dave Taht wrote: > > > *** Round 5 Plans > > * If I do another high RTT test, make rtt 1000ms default and try even higher > > * From Dave: tcp bbr, cdg, reno? dctcp would be weirdly interesting. > > * From Dave: slot 4ms 4ms bytes 10k 16 > > * If I get time I could change flenter to do asymmetric bandwidth tests and > go > back to Dave’s four box config with ingress cake, which would probably be a > more > common config. All depends on time available… > > > Well, let's see what happens on netdev. > > I took a few steps towards ripping out blue for comparison the other > day. > > > Ok, irtt needs shoring up before its first “pre-release release”, likely to > be in mid-December. I’ll focus on that with what time I have (which is less > now until the new year) until more is known. > > I think we can say Cake is stable and helps in many situations over what’s > currently available. There are smaller mysteries in different corners, but > they usually come down to something like: > > - the queue has been lost > - it's configured wrong > - that’s how it should be > - the device or driver isn’t working right > - the device is multi-queue > - the cpu is over-taxed > - the timer resolution isn’t high enough > > Maybe we can make these deployment issues easier for people to deal with > over time, first through documentation, then proposing and implementing > solutions if they become clear. Overall though, I’m glad Cake is available, > because in many situations people can just add a few keywords and enjoy the > benefits… "sch_cake cookbook" > > ___ > Cake mailing list > Cake@lists.bufferbloat.net > https://lists.bufferbloat.net/listinfo/cake > -- Dave Täht CEO, TekLibre, LLC http://www.teklibre.com Tel: 1-669-226-2619 ___ Cake mailing list Cake@lists.bufferbloat.net https://lists.bufferbloat.net/listinfo/cake
Re: [Cake] cake flenter results round 4
> On Dec 4, 2017, at 4:29 AM, Dave Tahtwrote: >> >> *** Round 5 Plans >> >> * If I do another high RTT test, make rtt 1000ms default and try even higher >> >> * From Dave: tcp bbr, cdg, reno? dctcp would be weirdly interesting. >> >> * From Dave: slot 4ms 4ms bytes 10k 16 >> >> * If I get time I could change flenter to do asymmetric bandwidth tests and >> go >> back to Dave’s four box config with ingress cake, which would probably be a >> more >> common config. All depends on time available… > > Well, let's see what happens on netdev. > > I took a few steps towards ripping out blue for comparison the other > day. Ok, irtt needs shoring up before its first “pre-release release”, likely to be in mid-December. I’ll focus on that with what time I have (which is less now until the new year) until more is known. I think we can say Cake is stable and helps in many situations over what’s currently available. There are smaller mysteries in different corners, but they usually come down to something like: - the queue has been lost - it's configured wrong - that’s how it should be - the device or driver isn’t working right - the device is multi-queue - the cpu is over-taxed - the timer resolution isn’t high enough Maybe we can make these deployment issues easier for people to deal with over time, first through documentation, then proposing and implementing solutions if they become clear. Overall though, I’m glad Cake is available, because in many situations people can just add a few keywords and enjoy the benefits… ___ Cake mailing list Cake@lists.bufferbloat.net https://lists.bufferbloat.net/listinfo/cake
Re: [Cake] cake flenter results round 4
Pete Heistwrites: > http://www.drhleny.cz/bufferbloat/cake/round4/ > > Round 4 Tarball: http://www.drhleny.cz/bufferbloat/cake/round4.tgz > > *** Notes/Analysis *** > > * I took the average satellite Internet latency of 638ms > (https://en.wikipedia.org/wiki/Satellite_Internet_access) and tried 20/20mbit > 320ms delay each direction. So here we have 20mbit symmetric Internet with > Cake > running on both the CPE and satellite- common config! I’m sorry we don’t have > Tooway 6/22mbit anymore for a real test, which was also here in desperation at > the house at some point. > > * Since I left the default rtt setting for most tests, these tests are an > exploration of what problems that can cause. I enjoy failed experiments though > (I’m looking at you penicillin). Here we can see that as we increase Cake’s > rtt > setting, total bandwidth improves, generally: > > http://www.drhleny.cz/bufferbloat/cake/round4/cake_rtt_200ms_rrulbe_eg_cake_18.0mbit/index.html > (22.77mbit) > http://www.drhleny.cz/bufferbloat/cake/round4/cake_rtt_400ms_rrulbe_eg_cake_18.0mbit/index.html > (26.17mbit) > http://www.drhleny.cz/bufferbloat/cake/round4/cake_rtt_600ms_rrulbe_eg_cake_18.0mbit/index.html > (25.99mbit) > http://www.drhleny.cz/bufferbloat/cake/round4/cake_rtt_800ms_rrulbe_eg_cake_18.0mbit/index.html > (25.96mbit) > http://www.drhleny.cz/bufferbloat/cake/round4/cake_rtt_1000ms_rrulbe_eg_cake_18.0mbit/index.html > (29.04mbit) > > * Cool. In this case, increasing rtt actually _improves_ host fairness, as > opposed to what we see with Ethernet around rtt 1ms, although TCP RTT suffers: Hmm. > > http://www.drhleny.cz/bufferbloat/cake/round4/hostiso_cake_200ms_eg_cake_src_cake_dst_18.0mbit/index.html > (1.83) > http://www.drhleny.cz/bufferbloat/cake/round4/hostiso_cake_400ms_eg_cake_src_cake_dst_18.0mbit/index.html > (1.64) > http://www.drhleny.cz/bufferbloat/cake/round4/hostiso_cake_600ms_eg_cake_src_cake_dst_18.0mbit/index.html > (1.57) > http://www.drhleny.cz/bufferbloat/cake/round4/hostiso_cake_800ms_eg_cake_src_cake_dst_18.0mbit/index.html > (1.45) > http://www.drhleny.cz/bufferbloat/cake/round4/hostiso_cake_1000ms_eg_cake_src_cake_dst_18.0mbit/index.html > (1.06) > > * Sorry for the delay and not testing much in the last few days. R.I.P. our > 16yo > European housecat who left us yesterday. I really liked that cat. I’ve enjoyed > looking at some of George’s results meanwhile... :( I used to have a cat, 'cindi clawford'. Been travelling too much to ever settle down again, it seems. > > *** Round 5 Plans > > * If I do another high RTT test, make rtt 1000ms default and try even higher > > * From Dave: tcp bbr, cdg, reno? dctcp would be weirdly interesting. > > * From Dave: slot 4ms 4ms bytes 10k 16 > > * If I get time I could change flenter to do asymmetric bandwidth tests and go > back to Dave’s four box config with ingress cake, which would probably be a > more > common config. All depends on time available… Well, let's see what happens on netdev. I took a few steps towards ripping out blue for comparison the other day. > > > ___ > Cake mailing list > Cake@lists.bufferbloat.net > https://lists.bufferbloat.net/listinfo/cake ___ Cake mailing list Cake@lists.bufferbloat.net https://lists.bufferbloat.net/listinfo/cake
Re: [Cake] cake flenter results round 4
On Sat, Dec 2, 2017 at 10:15 PM Jonathan Mortonwrote: > Just to note, deficiencies in host fairness are most likely directly > linked to less-than-full throughput. One host is able to use bandwidth > left unused by the other. > I see, that makes sense. Anyway what I wrote earlier about this high rtt test being opposite from Ethernet made no sense. :| It’s the same here as earlier, that increasing the rtt setting improves fairness, only there may be different things going on at these rtts. Thanks... ___ Cake mailing list Cake@lists.bufferbloat.net https://lists.bufferbloat.net/listinfo/cake
Re: [Cake] cake flenter results round 4
Just to note, deficiencies in host fairness are most likely directly linked to less-than-full throughput. One host is able to use bandwidth left unused by the other. - Jonathan Morton ___ Cake mailing list Cake@lists.bufferbloat.net https://lists.bufferbloat.net/listinfo/cake