netdev - Re: tcp_bbr: Forcing set of BBR congestion control as default

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 2 Jan 2017 14:12:16 -0500
From:   Neal Cardwell <ncardwell@...gle.com>
To:     sedat.dilek@...il.com
Cc:     Netdev <netdev@...r.kernel.org>
Subject: Re: tcp_bbr: Forcing set of BBR congestion control as default

On Mon, Jan 2, 2017 at 1:49 PM, Sedat Dilek <sedat.dilek@...il.com> wrote:
> On Mon, Jan 2, 2017 at 7:17 PM, Neal Cardwell <ncardwell@...gle.com> wrote:
>> On Mon, Jan 2, 2017 at 12:05 AM, Sedat Dilek <sedat.dilek@...il.com> wrote:
>>>
>>> Hi,
>>>
>>> I am trying to force the set of BBR congestion control as default.
>>> My old linux-config uses CUBIC as default.
>>> I want both BBR and CUBIC to be built but BBR shall be my default.
>>>
>>> I tried the below snippet.
>>>
>>> I refresh my new linux-config like this...
>>>
>>> $ MAKE="make V=1" ; COMPILER="mycompiler" ; MAKE_OPTS="CC=$COMPILER
>>> HOSTCC=$COMPILER" ; yes "" | $MAKE $MAKE_OPTS oldconfig && $MAKE
>>> $MAKE_OPTS silentoldconfig < /dev/null
>>>
>>> What am I doing wrong?
>>
>> Perhaps your build directory already has an old .config file that sets
>> the DEFAULT_TCP_CONG to be "cubic", and your "make oldconfig" and
>> "make silentoldconfig" are maintaining those lines from the old
>> .config?
>>
>> If you want to start with your existing .config but then make BBR the
>> default, then it might be simplest just to edit your .config directly:
>>
>> -CONFIG_DEFAULT_TCP_CONG="cubic"
>> +CONFIG_DEFAULT_TCP_CONG="bbr"
>>
>
> Just to clarify...
>
> I can have both TCP_CONG cubic and bbr built and switch for example via sysctl?

Yes, you can set the default congestion control using sysctl, eg:

  sysctl net.ipv4.tcp_congestion_control=bbr

And (as mentioned in the quick-start guide) on many Linux systems you
can make this "sticky" after reboots with something like:

  sudo bash -c 'echo "net.ipv4.tcp_congestion_control=bbr" >> /etc/sysctl.conf'

> Which tc version is required?
> Here tc is from iproute (20121211-2~precise).
> Is that enough?

You should not need a particular version of tc to install "fq".

In the BBR quick-start guide we recommend setting fq to be the default qdisc:

  sudo bash -c 'echo "net.core.default_qdisc=fq" >> /etc/sysctl.conf'

But if that doesn't work in your environment, then I believe you
should be able to install the fq qdisc with any version of tc, with
something like:

  tc qdisc replace dev eth0 root fq

You just won't be able to set or view configuration options.

>> BTW, I presume you have seen the warning in the BBR commit message or
>> tcp_bbr.c about ensuring that BBR is used with the "fq" qdisc:
>>
>>   NOTE: BBR *must* be used with the fq qdisc ("man tc-fq") with pacing
>>   enabled, since pacing is integral to the BBR design and
>>   implementation. BBR without pacing would not function properly, and
>>   may incur unnecessary high packet loss rates.
>>
>> The BBR quick-start guide has some details about how to build and
>> enable BBR and fq:
>>
>>   https://github.com/google/bbr/blob/master/Documentation/bbr-quick-start.md
>>
>
> Hmmm...
>
> From [1] Section "Further reading"...
>
> egrep '(CONFIG_TCP_CONG_BBR|CONFIG_NET_SCH_FQ)=' .config
>
> then you see exactly the following lines:
>
> CONFIG_TCP_CONG_BBR=y
> CONFIG_NET_SCH_FQ=y
>
> Should CONFIG_TCP_CONG_BBR have a "select CONFIG_NET_SCH_FQ" in its Kconfig?
> That would be safer.
>
> [1] https://github.com/google/bbr/blob/master/Documentation/bbr-quick-start.md

That would be a little safer, but not sufficient (since the qdisc
still has to be configured to be in the transmit path somewhere).

thanks,
neal