[Codel] fq_Codel vs Sfq_Codel

[Codel] fq_Codel vs Sfq_Codel

[divya singla]

[-- Attachment #1: Type: text/plain, Size: 76 bytes --]

can anybody tell me is there any difference between fq_Codel and Sfq_Codel?

[-- Attachment #2: Type: text/html, Size: 138 bytes --]

Re: [Codel] fq_Codel vs Sfq_Codel

[Dave Taht]

sfq_codel in ns2 is derived more from SFQ than DRR, and there are
several substantial differences, although sfq_codel  can be almost,
but not quite, configured to act the same as the dual DRR version that
is in the actual, shipping, fq_codel. The version of DRR fq_codel uses
is more similar to "DRR+" and "DRR++", but is uniquely it's own.

You have to use sfq_codel VERY carefully in the ns2 version, as the
ability to correctly generate a five tuple hash is not there.
Correctly generating a 5 tuple is almost, but not quite, in the ns3
version, last I looked, and that ns3 version is MUCH closer to the
actual fq_codel version deployed in the field, but still not quite the
same thing.  I have kind of lost track as to the right ns2 and ns3
versions to start from, also. Tom?

in terms of setting up realistic scenarios: A common (actually, nearly
universal) change to fq_codel in the field at low bandwidths (below
50mbit) is to use a reduced quantum (300 bytes). I don't remember how
to do that right in the ns2 version. It has, after all, been 3 years
of no interest from the academic community. Also the codel target and
interval need to be tuned slightly, at below 2.5mbits to account for a
single full length packet at those speeds.

I am delighted by the sudden upsurge in interest in the *fq_codel
versions but I would certainly appreciate more efforts on making the
underlying simulations actually match the deployment, before everyone
goes off writing papers again.

See sch_fq_codel.c in the linux distributions since 3.6, see also
codel.h there, and compare against what you have for ns2 and ns3.

I completely lack time or funding to make sims match the (much better
reality) at this point. Worst case, I would encourage you to test the
sims verses the deployed code with the same exact tests, so you can
see what differences exist.


On Sat, Feb 28, 2015 at 9:34 AM, divya singla <divyasingla1989@gmail.com> wrote:
> can anybody tell me is there any difference between fq_Codel and Sfq_Codel?
>
>
>
> _______________________________________________
> Codel mailing list
> Codel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel
>



-- 
Dave Täht
Let's make wifi fast, less jittery and reliable again!

https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb

Re: [Codel] fq_Codel vs Sfq_Codel

[Dave Taht]

ah, it looks like tom is making some progress on ns3 here:

http://code.nsnam.org/tomh/ns-3-dev-aqm

On Sat, Feb 28, 2015 at 9:58 AM, Dave Taht <dave.taht@gmail.com> wrote:
> sfq_codel in ns2 is derived more from SFQ than DRR, and there are
> several substantial differences, although sfq_codel  can be almost,
> but not quite, configured to act the same as the dual DRR version that
> is in the actual, shipping, fq_codel. The version of DRR fq_codel uses
> is more similar to "DRR+" and "DRR++", but is uniquely it's own.
>
> You have to use sfq_codel VERY carefully in the ns2 version, as the
> ability to correctly generate a five tuple hash is not there.
> Correctly generating a 5 tuple is almost, but not quite, in the ns3
> version, last I looked, and that ns3 version is MUCH closer to the
> actual fq_codel version deployed in the field, but still not quite the
> same thing.  I have kind of lost track as to the right ns2 and ns3
> versions to start from, also. Tom?
>
> in terms of setting up realistic scenarios: A common (actually, nearly
> universal) change to fq_codel in the field at low bandwidths (below
> 50mbit) is to use a reduced quantum (300 bytes). I don't remember how
> to do that right in the ns2 version. It has, after all, been 3 years
> of no interest from the academic community. Also the codel target and
> interval need to be tuned slightly, at below 2.5mbits to account for a
> single full length packet at those speeds.
>
> I am delighted by the sudden upsurge in interest in the *fq_codel
> versions but I would certainly appreciate more efforts on making the
> underlying simulations actually match the deployment, before everyone
> goes off writing papers again.
>
> See sch_fq_codel.c in the linux distributions since 3.6, see also
> codel.h there, and compare against what you have for ns2 and ns3.
>
> I completely lack time or funding to make sims match the (much better
> reality) at this point. Worst case, I would encourage you to test the
> sims verses the deployed code with the same exact tests, so you can
> see what differences exist.
>
>
> On Sat, Feb 28, 2015 at 9:34 AM, divya singla <divyasingla1989@gmail.com> wrote:
>> can anybody tell me is there any difference between fq_Codel and Sfq_Codel?
>>
>>
>>
>> _______________________________________________
>> Codel mailing list
>> Codel@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/codel
>>
>
>
>
> --
> Dave Täht
> Let's make wifi fast, less jittery and reliable again!
>
> https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb



-- 
Dave Täht
Let's make wifi fast, less jittery and reliable again!

https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb

Re: [Codel] fq_Codel vs Sfq_Codel

[Tom Henderson]

On 02/28/2015 09:58 AM, Dave Taht wrote:
> sfq_codel in ns2 is derived more from SFQ than DRR, and there are
> several substantial differences, although sfq_codel  can be almost,
> but not quite, configured to act the same as the dual DRR version that
> is in the actual, shipping, fq_codel. The version of DRR fq_codel uses
> is more similar to "DRR+" and "DRR++", but is uniquely it's own.
>
> You have to use sfq_codel VERY carefully in the ns2 version, as the
> ability to correctly generate a five tuple hash is not there.
> Correctly generating a 5 tuple is almost, but not quite, in the ns3
> version, last I looked, and that ns3 version is MUCH closer to the
> actual fq_codel version deployed in the field, but still not quite the
> same thing.  I have kind of lost track as to the right ns2 and ns3
> versions to start from, also. Tom?

Dave and all,

The status of ns-2 and ns-3 support is as follows.

In ns-2, some new models (CoDel, SFQCoDel, CoDel-DT, and PIE) 
contributed by Kathleen Nichols, CableLabs, and Cisco will be part of 
the ns-2.36 release, which is in preparation.  An initial release 
candidate was posted here:
http://nsnam.isi.edu/nsnam/index.php/Roadmap#Release_candidates_for_testing

and the code in the CVS tree is consistent with this.  (Preview) 
documentation on the new models is available here:
http://www.isi.edu/nsnam/dist/release/rc1/doc/node69.html
and here:
http://www.isi.edu/nsnam/dist/release/rc1/doc/node89.html

When will 2.36 be released?  I am working through a long list of issues 
getting the debug build to work on clang-based systems, and when those 
are resolved (in the next few weeks?) I'll release 2.36.  The release 
candidate above is pretty close to what will be the final bits.

In ns-3, basic CoDel was added in n-3.21.  Dave and Andrew McGregor 
contributed some models for FQ- and SFQ-CoDel a while back; links to 
this code can be found on our wiki Current Development page.  Inclusion 
of these is blocked by some stack refactoring that Dave pointed to in a 
subsequent post, to insert a priority queue sublayer for these queues. 
I'm working towards the goal of making these available by ns-3.23 (May), 
but I aimed for ns-3.22 and missed, so I can't promise that these will 
be done by May; we'll see.

If anyone wants to help test these new simulation models and get them 
into shape for release, please contact me.

- Tom