Thanks very much for the information!  (I've been reading 20+ page threads on 
Ubiquity's forum looking for messages that you've posted)

I'm going to be testing an EdgeRouter X which doesn't support hardware 
acceleration so I don't think I have to worry about TCP offload (?).  (I'd like 
to use my existing x86 hardware but I can't find a solid router OS that I like)

I'll also go ahead and enable fq_codel on my downstream and hope for the best.

Thanks again,
John


> On Feb 16, 2016, at 2:19 PM, Dave Taht <[email protected]> wrote:
> 
>> On Tue, Feb 16, 2016 at 10:06 AM, John Klimek <[email protected]> wrote:
>> Does anybody have any suggested quantum/target values for a cable modem 
>> connection of 25/3?  (25 Mbps down, 3 Mbps up)
> 
> I would use a quantum 300 for the up. There is no need to fiddle with
> the target values below 2.5mbits.
> 
> The edgerouter has issues with tcp offloads enabled. If you are not
> using the other ports as routed ports, you can disable gro offloads on
> all ports for lower latency at a huge cost in local bandwidth (this is
> something that cake will fix if we can get ubnt to ship it one day.
> http://www.bufferbloat.net/projects/codel/wiki/CakeTechnical )
> 
> 
>> 
>> It looks like the defaults for Ubiquity EdgeRouter should work but I'm not 
>> sure if I can achieve better performance by tuning anything?
> 
> Make a change, measure with a good benchmark like flent's rrul test.
> 
>> Also, should I enable fq_codel on my downstream or only my upstream?
> 
> Both are severely overbuffered on cablemodems and CMTSes running at this 
> speed.
> 
> http://burntchrome.blogspot.com/2014/05/fixing-bufferbloat-on-comcasts-blast.html
> 
> https://wiki.openwrt.org/doc/howto/sqm
> 
> 
>> Thanks!
>> 
>> - John
>> _______________________________________________
>> Bloat mailing list
>> [email protected]
>> https://lists.bufferbloat.net/listinfo/bloat
_______________________________________________
Bloat mailing list
[email protected]
https://lists.bufferbloat.net/listinfo/bloat

Reply via email to