Re: Priorizing empty ACKs
On Fri, 28 Feb 2003 10:21:55 +0100 Daniel Hartmeier <[EMAIL PROTECTED]> wrote: >On Thu, Feb 27, 2003 at 10:13:55PM -0800, Ben Lovett wrote: > >> All in all, I'm seeing a great improvement. My connection is ADSL >> 1.5M/384, and the sweet spot for my connection appears to be 330Kbit/s. >> I'll do some more playing around with it tomorrow to see if I can get >> better speeds, but even if I don't -- I'm happy with what I'm getting >> now :) > >Great, thanks for the feedback. > >Henning suggested using the priq scheduler instead of cbq, like > > altq on kue0 priq bandwidth 100Kb queue { q_pri, q_def } > queue q_pri priority 7 > queue q_def priority 1 priq(default) > >I get about the same results with this, but it's somewhat simpler. >BTW, try pfctl -vsq or -vvsq to check the altq stats. > >Daniel This is awesome :)) this way my girlfriend can use my connection while me and my friends are sharing data :) Is this patch going to be in 3.3 ?? I'm running 28/2/2003 3.2-current and this patch did the job. Would be nice to see it in 3.3 so I can roll out it easily on wrk servers. Michiel van Baak
Re: Priorizing empty ACKs
On Thu, Feb 27, 2003 at 10:13:55PM -0800, Ben Lovett wrote: > All in all, I'm seeing a great improvement. My connection is ADSL > 1.5M/384, and the sweet spot for my connection appears to be 330Kbit/s. > I'll do some more playing around with it tomorrow to see if I can get > better speeds, but even if I don't -- I'm happy with what I'm getting > now :) Great, thanks for the feedback. Henning suggested using the priq scheduler instead of cbq, like altq on kue0 priq bandwidth 100Kb queue { q_pri, q_def } queue q_pri priority 7 queue q_def priority 1 priq(default) I get about the same results with this, but it's somewhat simpler. BTW, try pfctl -vsq or -vvsq to check the altq stats. Daniel
Re: Priorizing empty ACKs
On Wednesday, February 26, 2003, at 03:31 PM, Daniel Hartmeier wrote: Let me know if this works as well for you. :) All in all, I'm seeing a great improvement. My connection is ADSL 1.5M/384, and the sweet spot for my connection appears to be 330Kbit/s. I'll do some more playing around with it tomorrow to see if I can get better speeds, but even if I don't -- I'm happy with what I'm getting now :) Ben -- Walk softly and carry a megawatt laser.
Re: Priorizing empty ACKs
If this works I will be terribly happy, until now I thought there was no way arround it. THANKYOU I will try implementing it tonight! scott
Re: Priorizing empty ACKs
On Thu, Feb 27, 2003 at 04:42:41AM +0100, Daniel Hartmeier wrote: > 5% -> 50% is quite good, I wonder if you can reach 95% with the right > settings. :) It looks like my cablemodem (no pppoe) upstream went from 128 -> 256 sometime without my ISP telling me or me noticing. I had initially used your value of 100. I tried gradually increasing it, and 224kb seems to be the sweet spot for my setup: 224 -> ~185 240 -> ~155 256 -> ~65 - jolan
Re: Priorizing empty ACKs
On Wed, Feb 26, 2003 at 09:28:05PM -0600, Jolan Luff wrote: > n n 209k/s > n y 11.2k/s > y n 209k/s > y y 103k/s > > Almost an order of magnitude faster... I'd say it works well :P I noticed some effects that I can't fully explain yet. If I set the altq bandwidth to 128Kb, download drops worse than when I set it somewhat lower (100Kb works best). I assume that's because the real upstream capacity is lowered by PPPoE, and giving altq a value higher than the real limit causes the priorizing to get less than optimal. If I lower the bandwidth value further, the download drop gets worse, too. So I guess you have to find the optimal value manually to get the best effect. 5% -> 50% is quite good, I wonder if you can reach 95% with the right settings. :) Daniel
Re: Priorizing empty ACKs
On Thu, Feb 27, 2003 at 12:31:42AM +0100, Daniel Hartmeier wrote: > Let me know if this works as well for you. :) I've been drooling over this ever since you mentioned it weeks/months ago. I tested this by grabbing a 26 meg file from kernel.org. Line saturation via someone scp'ing a file from my machine to theirs. altqsaturated speed n n 209k/s n y 11.2k/s y n 209k/s y y 103k/s Almost an order of magnitude faster... I'd say it works well :P - jolan