[Cake] fq in p4

[Cake] fq in p4

[Dave Taht]

https://homes.cs.washington.edu/~arvind/papers/afq.pdf

-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619

Re: [Cake] fq in p4

[Stephen Hemminger]

[-- Attachment #1: Type: text/plain, Size: 456 bytes --]

Typical academic paper testing with Reno.
At least they compared DCTCP

On Wed, Jul 25, 2018, 10:04 PM Dave Taht <dave.taht@gmail.com> wrote:

> https://homes.cs.washington.edu/~arvind/papers/afq.pdf
>
> --
>
> Dave Täht
> CEO, TekLibre, LLC
> http://www.teklibre.com
> Tel: 1-669-226-2619
> _______________________________________________
> Cake mailing list
> Cake@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cake
>

[-- Attachment #2: Type: text/html, Size: 1116 bytes --]

Re: [Cake] fq in p4

[Dave Taht]

well, they got a working in-hardware prototype, and a p4
implementation, it uses DRR, shows feasability
if nothing else. (they also report big wins throughout, and their aqm
works against flow isolation).

Happy. I asked them if they had any code we can play with.

On Thu, Jul 26, 2018 at 11:17 AM Stephen Hemminger
<stephen@networkplumber.org> wrote:
>
> Typical academic paper testing with Reno.
> At least they compared DCTCP
>
> On Wed, Jul 25, 2018, 10:04 PM Dave Taht <dave.taht@gmail.com> wrote:
>>
>> https://homes.cs.washington.edu/~arvind/papers/afq.pdf
>>
>> --
>>
>> Dave Täht
>> CEO, TekLibre, LLC
>> http://www.teklibre.com
>> Tel: 1-669-226-2619
>> _______________________________________________
>> Cake mailing list
>> Cake@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/cake



-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619

Re: [Cake] fq in p4

[Dave Taht]

some word from avind's group.

"Thanks Dave for the pointer!  Sounds very interesting, so we will
definitely take a look.

We do have p4 code for AFQ and happy to send it your way.  There are
two crucial components to the AFQ implementation -- one is the switch
state stuff and that is easily expressed in p4, and the other is the
queue scheduling stuff which p4 currently doesn't provide any
knobs/mechanisms for realizing it.  But there are some upcoming
Barefoot mechanisms that can help us directly realize the necessary
mechanism (i.e., the rotating priority queue mechanism as described in
the paper) and we do think that the queue scheduling disciplines would
eventually make its way into standard p4 in some time..."

and

"Thanks for all of the pointers, especially the VJ talk!  We are in
fact working on timer wheels and calendar queues as a scheduling
primitive to ensure timely delivery of messages at the application
level -- for example, we can use such mechanisms to ensure low tail
latency for key-value requests and/or map-reduce co-flows.

We will get you the code and do go ahead and forward our work/emails
to the mailing lists."

my editorial comment: JOY. BLISS. RAPTURE. There is light at the end
of this tunnel.





On Thu, Jul 26, 2018 at 11:33 AM Dave Taht <dave.taht@gmail.com> wrote:
>
> well, they got a working in-hardware prototype, and a p4
> implementation, it uses DRR, shows feasability
> if nothing else. (they also report big wins throughout, and their aqm
> works against flow isolation).
>
> Happy. I asked them if they had any code we can play with.
>
> On Thu, Jul 26, 2018 at 11:17 AM Stephen Hemminger
> <stephen@networkplumber.org> wrote:
> >
> > Typical academic paper testing with Reno.
> > At least they compared DCTCP
> >
> > On Wed, Jul 25, 2018, 10:04 PM Dave Taht <dave.taht@gmail.com> wrote:
> >>
> >> https://homes.cs.washington.edu/~arvind/papers/afq.pdf
> >>
> >> --
> >>
> >> Dave Täht
> >> CEO, TekLibre, LLC
> >> http://www.teklibre.com
> >> Tel: 1-669-226-2619
> >> _______________________________________________
> >> Cake mailing list
> >> Cake@lists.bufferbloat.net
> >> https://lists.bufferbloat.net/listinfo/cake
>
>
>
> --
>
> Dave Täht
> CEO, TekLibre, LLC
> http://www.teklibre.com
> Tel: 1-669-226-2619



--

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619

Re: [Cake] fq in p4

[Toke Høiland-Jørgensen]

Dave Taht <dave.taht@gmail.com> writes:

> some word from avind's group.
>
> "Thanks Dave for the pointer!  Sounds very interesting, so we will
> definitely take a look.
>
> We do have p4 code for AFQ and happy to send it your way. There are
> two crucial components to the AFQ implementation -- one is the switch
> state stuff and that is easily expressed in p4, and the other is the
> queue scheduling stuff which p4 currently doesn't provide any
> knobs/mechanisms for realizing it. But there are some upcoming
> Barefoot mechanisms that can help us directly realize the necessary
> mechanism (i.e., the rotating priority queue mechanism as described in
> the paper) and we do think that the queue scheduling disciplines would
> eventually make its way into standard p4 in some time..."

Yeah, the lack of any interface to the queueing in p4 always bothered
be. Good to see that is (hopefully) being worked on :)

-Toke