[Ecn-sane] [bbr-dev] Applying BBRv2 patches

Dave Taht dave.taht at gmail.com
Mon Mar 18 05:26:12 EDT 2019


On Thu, Mar 14, 2019 at 11:21 AM Jared Newman <jnewman2501 at gmail.com> wrote:
>
> Hey Neal,
>
> Just a quick update: turns out the 1-2 second delays we were seeing were because we had a net.ipv4.tcp_notsent_lowat of 4MB (!!). Setting this to 16KB made BBR perform beautifully on the link, and reduced all queue delays to within the expected RTT variance. I'm still keen on comparing BBR with and without these patches, so will report back shortly with these results.

I am a huge fan of this option, (apple has it too) and certainly the
default is far too high.

I would really love to have an equivalent net.ipv4.tcp_notsent_lowat
for udp sockets as well as a setsockopt. Has that happened yet?


> Cheers,
> Jared.
>
> On Tuesday, March 12, 2019 at 10:50:40 PM UTC-7, Neal Cardwell wrote:
>>
>> Thanks for the report! We'd be interested to hear how it goes.
>>
>> thanks,
>> neal
>>
>> On Tue, Mar 12, 2019 at 10:44 PM Jared Newman <jnewm... at gmail.com> wrote:
>>>
>>> Ah awesome, glad to see the improvements make it into the next kernel! Thanks for the pointers and quick response, Neal.
>>>
>>> We're running TCP BBR on a 2.4GHz mobile ad hoc network and are seeing an excessive outgoing queue build up, leading to application delays of 1-2 seconds on a link that has a BBR-estimated min-RTT of ~15ms. By the sounds of it, feature (2) will be what we want so we'll try applying the patches on top of 18845557fd6f and see how things go.
>>>
>>> Cheers,
>>> Jared.
>>>
>>> On Tuesday, March 12, 2019 at 10:29:08 PM UTC-7, Neal Cardwell wrote:
>>>>
>>>> Hi,
>>>>
>>>> For the patches in  https://groups.google.com/forum/#!topic/bbr-dev/8pgyOyUavvY there were two main features:
>>>>
>>>> (1) Higher throughput for wifi and other paths with aggregation
>>>>
>>>> (2) Lower queuing delays by frequently draining excess in-flight data
>>>>
>>>>
>>>> For the patches related to (1), that functionality is already in davem/net-next and torvalds/linux, as of recently, and those patches will be in Linux 5.1. We were waiting until Linux 5.1 is released to announce that on this e-mail list. But since it came up anyway, we may as well mention it. ;-)
>>>>
>>>>
>>>> For the patches related to (2), that functionality is still undergoing testing on YouTube and elsewhere.
>>>>
>>>>
>>>> If you want to apply the original patches from  https://groups.google.com/forum/#!topic/bbr-dev/8pgyOyUavvY  then you can apply them on top of the Linux commit:
>>>>
>>>> 18845557fd6f
>>>>
>>>> from March 2018. That's where they were originally generated.
>>>>
>>>>
>>>> best,
>>>>
>>>> neal
>>>>
>>>>
>>>>
>>>> On Tue, Mar 12, 2019 at 10:09 PM Jared Newman <jnewm... at gmail.com> wrote:
>>>>>
>>>>> Hey BBR-dev,
>>>>>
>>>>> I'm trying to apply the patches given in https://groups.google.com/forum/#!topic/bbr-dev/8pgyOyUavvY, but am struggling to find what commit of next-net I should apply them on. I tried a couple different tags in https://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git but would always get conflicts that I didn't feel comfortable resolving. Does anyone have a suggestion / commit hash to apply them to?
>>>>>
>>>>> Cheers,
>>>>> Jared.
>>>>>
>>>>> --
>>>>> You received this message because you are subscribed to the Google Groups "BBR Development" group.
>>>>> To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u... at googlegroups.com.
>>>>> For more options, visit https://groups.google.com/d/optout.
>>>
>>> --
>>> You received this message because you are subscribed to the Google Groups "BBR Development" group.
>>> To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u... at googlegroups.com.
>>> For more options, visit https://groups.google.com/d/optout.
>
> --
> You received this message because you are subscribed to the Google Groups "BBR Development" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+unsubscribe at googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.



-- 

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740


More information about the Ecn-sane mailing list