Re: Using Click RawSocket on PlanetLab

Dimitris,

Thanks for providing this patch!  Would you mind creating a Github pull 
request for it?

Best,
Eddie


On 8/26/11 5:28 AM, Dimitris Syrivelis wrote:
>
>   Hello all,
>
>   We attempted to deploy a click configuration on planetlab. We have tested both
> RawSocket as well as IPFlowRawSockets and AggregateIPFLows combo as it is
> suggested. Unfortunately, the receiving side does not work. We have tested on
> PlanetLab Europe (PLE) nodes with latest planetlab linux distro that runs over
> Xen or Vserver.
>
>   Running rawsockets on planetlab is a dominant planetlab list issue with several
> threads and different solutions that are functional for specific planetlab
> deployment versions. Starting with Proper API which is the oldest and probably
> deprecated, Scout kernel module and currently vmnet.
>   We found that the solution that currently applies on latest PLE nodes is
> described on this thread:
> http://lists.planet-lab.org/pipermail/devel/2009-January/003255.html.
>
>   In brief, the authors of vmnet+ explain that in order for your sliver to become
> a port owner on a planetlab machine you need to create and bind a DGRAM or
> STREAM socket for UDP or TCP respectively. This will trigger the port
> registration code in vmnet+. In the first version of vmnet, binding the raw
> socket port was adequate and this is what current git rawsockets.cc code does.
>
>   Attached you will find a patch for rawsockets.cc,hh that creates a seperate
> standard socket and binds it to the port to achieve registration. We have also
> removed proper support since it is obsolete on planetlab. We have tested this
> code sending/receiving UDP traffic on PLE nodes. Maybe we need other users to
> confirm that it runs on PLC before it is considered for integration with
> mainstream click.
>
>   Regards,
>
>    Dimitris
>
>
>
> _______________________________________________
> click mailing list
> [email protected]
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click