[Cake] Cake over satellite
moeller0 at gmx.de
Mon Oct 19 15:58:25 EDT 2015
On October 19, 2015 10:48:26 AM CDT, Kevin Darbyshire-Bryant <kevin at darbyshire-bryant.me.uk> wrote:
>On 19/10/15 16:15, Sebastian Moeller wrote:
>> Hi Noven,
>> On Oct 19, 2015, at 09:55 , Noven Purnell-Webb
><noven at liberant.com.au> wrote:
>>> I'm about to go on a mission to the outback, with one objective
>being to tune a disappointing satellite link. Theoretically 6/1mbps,
>with RTTs ranging from 700 - 1200ms. From the cake technical page
>>> "The AQM layer has no configuration options. However, it is planned
>to add simple tuning options for different prevailing RTTs that may be
>significantly different from the Internet-scale 100ms currently
>assumed. Satellite links tend to impose longer RTTs, and enclosed LANs
>tend to have much shorter RTTs." Does this imply cake is not yet ready
>for use on satellite links? Or merely that I'll get sub-optimal results
>but it's still worth testing?
>>> I see I could set the interval in sch_cake.c and recompile - is
>there any reason why this would be a bad idea? Am I right in thinking
>with such observed variation I'll get better results setting for the
>lower limit (700) rather than the upper (1200)?
>>> Any suggestions for a more sane target to set while I'm doing this?
>>> Has anyone here already done testing on satellite links?
>> I believe that interval/rtt is now settable via tc, BUT target is
>clamped to a max of 5ms, while in theory it should be 5-10% of rtt, so
>current cake might work well out of the box, or might require fiddling
>with target. It would be most excellent if you could actually test that
>aspect of cake for us ;)
>The interval value is exposed as an rtt parameter, either as a time
>700ms) or as some hopefully relevant pre-set keywords, may I suggest
>'satellite' is suitable in your case (1000ms) The target value is not
>directly available from tc, however an API interface for it does exist
>(like interval alias rtt and other parameters) Target max is 5ms,
>except for 'slow' links where a calculation using the time for an MTU
>sized packet + overhead is used (there's little point in trying to
>control latency to 5ms when it takes longer than that to actually send
>the bytes out), other than that target is set to 5% of rtt.
Since target max is clamped to 5ms, you effectively only get the 5% setting for rtt./interval below 100 ms. Which might or might not be correct for a long latency satellite link. All I want to emphasize is that somebody should test this ;)...
>Tell it the link bandwidth, any overhead bytes (PPP, ether-FCS, VLan
>the like) and the rtt 1000ms and she'll be right mate ;-)
>> Best Regards
>>> - Noven
>>> - Noven Purnell-Webb
>>> Creative Technology Consultant
>>> +61 448 841 091
>>> Cake mailing list
>>> Cake at lists.bufferbloat.net
>> Cake mailing list
>> Cake at lists.bufferbloat.net
>Cake mailing list
>Cake at lists.bufferbloat.net
Sent from my Android device with K-9 Mail. Please excuse my brevity.
More information about the Cake