[P4-dev] Setup ideal for experiments meant for SIGCOMM papers
andy.fingerhut at gmail.com
Mon Jul 17 19:30:59 EDT 2017
I am not aware of any "P4 fraternity". Just realize that there can be as
many, or more, reasons for getting different packet forwarding performance
across P4-programmable devices as there are for getting different program
execution performance across different CPU families (e.g. x86, ARM,
PowerPC, MIPS, etc.). I would expect the control plane update rate to
depend heavily not only on what the P4 target device was, and its
driver/support software, but also on the transport layer protocol (e.g.
protobufs, Thrift, etc.) and probably several other factors.
I would expect that whether your paper is accepted for publication would
have more to do with the novelty and expected impact of your ideas. For
example, if the primary content of the paper was "here is a P4 program I
wrote that does a subset of what the open source switch.p4 program does
, which has been published for years, and some measurements of control
plane update and packet forwarding rates, run on some open source programs
written and published by others  ", then I would expect most academic
reviewers would find that to have too little original content to be worth
If the primary content of the paper was "Here is a P4 program that
optimizes common application X by a factor of 10x, in a way that we have
never seen published before", then P4 becomes simply one implementation
approach for your original ideas, not the 'meat' of the paper, and they are
likely to look more favorably on it.
On Mon, Jul 17, 2017 at 9:37 AM, Rakesh Datta <rakesh.datta6 at gmail.com>
> Hello Andy,
> Thanks for your suggestion.
> Basically, I have a centralized P4 Controller which is controlling remote
> P4 switches.
> - I want to test the performance of the flow-rules push into the
> data-plane, from the Controller to the switch.
> - I also want to test the packet processing time of the P4-switch, which
> includes new tables and logic I have defined.
> So far, I have been using bvm2 and I have some good results. Qs - is this
> good enough experiment for a P4 paper?
> If yes, is it ok test it over VMs located on my laptop.
> If no, does the P4 fraternity always test their code on the hardware you
> mentioned, for the sake of the paper?
> Thanks 'n' Regards,
> Rakesh Datta
> +1 6692628019 <(669)%20262-8019> (USA)
> On Mon, Jul 17, 2017 at 7:59 AM, Andy Fingerhut <andy.fingerhut at gmail.com>
>> What do you intend to test via your P4 testing environment? Correctness
>> of the P4 programs? Performance of control plane updates? Performance of
>> packet forwarding?
>> If it is the performance of packet forwarding, then please realize that
>> while the behavioral-model code does have some amount of performance tuning
>> in it, the primary use case for it is debugging and testing of your P4
>> programs. There are NPU-based, FPGA-based, and ASIC-based P4-programmable
>> environments that can forward packets several orders of magnitude faster
>> than it can.
>> On Sun, Jul 16, 2017 at 7:51 PM, Rakesh Datta <rakesh.datta6 at gmail.com>
>>> Hello All,
>>> I am working on a P4 paper and planning to publish it in SIGCOMM or any
>>> top-tier conference.
>>> For this, my entire setup is based on Linux virtual machines running on
>>> my laptop.
>>> Can you please suggest what should be the ideal platform for setting up
>>> P4 testing environment targetted for a paper.
>>> Folks who have been actively working on SIGCOMM papers, can you please
>>> Thanks 'n' Regards,
>>> Rakesh Datta
>>> P4-dev mailing list
>>> P4-dev at lists.p4.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the P4-dev