[Bloat] Difference between codel and fq_codel?

[Bloat] Difference between codel and fq_codel?

[John Klimek]

I'm currently using pfaense which only supports codel and not fq_codel. Is there a big difference between them?  Is it worth looking into using a different router?

Sent from my iPhone

Re: [Bloat] Difference between codel and fq_codel?

[Rick Jones]

On 02/03/2016 11:22 AM, John Klimek wrote:
> I'm currently using pfaense which only supports codel and not
> fq_codel. Is there a big difference between them?  Is it worth
> looking into using a different router?

My simplistic understanding is fq_codel creates several/many different 
queues, spreading flows across those queues and applying codel on each 
queue.


http://man7.org/linux/man-pages/man8/tc-fq_codel.8.html

http://man7.org/linux/man-pages/man8/tc-codel.8.html

rick jones

Re: [Bloat] Difference between codel and fq_codel?

[Jonathan Morton]

> On 3 Feb, 2016, at 22:09, Rick Jones <rick.jones2@hpe.com> wrote:
> 
> On 02/03/2016 11:22 AM, John Klimek wrote:
>> I'm currently using pfaense which only supports codel and not
>> fq_codel. Is there a big difference between them?  Is it worth
>> looking into using a different router?
> 
> My simplistic understanding is fq_codel creates several/many different queues, spreading flows across those queues and applying codel on each queue.

There *is* a big difference between fq_codel and plain codel.  Simply put, the “fq” part of fq_codel is capable of mitigating inter-flow induced latency much more reliably than plain codel can.  Codel in itself manages only intra-flow induced latency.

Fq_codel maintains a separate codel instance per queue, so it also does a better job of applying the correct amount of congestion feedback to each flow, rather than a blanket amount over all traffic using the link.

The difference is particularly marked when you are dealing with congestion-unresponsive traffic (which codel is inherently poor at managing), but is measurable and even noticeable even with standard TCP flows.

 - Jonathan Morton

Re: [Bloat] Difference between codel and fq_codel?

[Jim Gettys]

[-- Attachment #1: Type: text/plain, Size: 1267 bytes --]

On Wed, Feb 3, 2016 at 3:09 PM, Rick Jones <rick.jones2@hpe.com> wrote:

> On 02/03/2016 11:22 AM, John Klimek wrote:
>
>> I'm currently using pfaense which only supports codel and not
>> fq_codel. Is there a big difference between them?  Is it worth
>> looking into using a different router?
>>
>
> My simplistic understanding is fq_codel creates several/many different
> queues, spreading flows across those queues and applying codel on each
> queue.
>
>
> http://man7.org/linux/man-pages/man8/tc-fq_codel.8.html
>
> http://man7.org/linux/man-pages/man8/tc-codel.8.html


​It also preferentially schedules "new" queues, relative to queues that
have a standing queue.  So short transactions such as dchp, ra, short RPC
transactions, VOIP, etc fly through the router relative to bulk TCP flows.

This algorithm does really wonderful things for any flow that is a "good
citizen" by pacing their traffic at their share of available bandwidth,
rather than bursting their traffic.  For the first time, there is incentive
to be a "good citizen" in your behavior.

>
>
> rick jones
>
> _______________________________________________
> Bloat mailing list
> Bloat@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat
>

[-- Attachment #2: Type: text/html, Size: 2647 bytes --]

Re: [Bloat] Difference between codel and fq_codel?

[Toke Høiland-Jørgensen]

John Klimek <jklimek@gmail.com> writes:

> I'm currently using pfaense which only supports codel and not
> fq_codel. Is there a big difference between them? Is it worth looking
> into using a different router?

Yes, and quite possibly. :)

Basically, Codel drops packets to try to keep queues small. This works
reasonably well. However, what FQ-CoDel adds is flow isolation and
priority for sparse flows, which is what in most cases gets rid of
almost all of the added latency under load.

For a (much) longer version of the above, see this:
http://www.sciencedirect.com/science/article/pii/S1389128615002479 -- In
particular, figure 2 shows the performance of a bunch of different
algorithms, including CoDel and FQ-CoDel :)

-Toke