[Planetlab-users] TCP raw Sockets in planet lab (AGAIN)

[Venkata Pingali]

Mark Huang wrote:
> Luis Carlos Erpen de Bona wrote:
>> Its start easy, on 10.0.0.1 capture the data going out to 10.0.0.2, 
>> send by a TCP tunnel for the correspondent real
>> IP.  At the target I'm supposed to deliver the data to
>> the application with source 10.0.0.1. But it's failing.
>> Bug in my code or planetlab feature?
>
> As Christoph mentioned, you cannot send packets with a non-local 
> source IP address, including RFC1918 addresses such as 10.0.0.0/8. 
> Even if you could, no router in its right mind would forward such 
> packets.
>
> Are you trying to build an overlay network? If so, the new TUN/TAP 
> feature of VNET, which will rollout to production nodes very shortly, 
> may help:
>
> http://www.planet-lab.org/doc/vnet.php
>
> In short, very soon, every PlanetLab node will have a virtual TUN/TAP 
> device "tap0" that will be configured with a unique (to each node) 
> RFC1918 10.0.0.0/8 address. An overlay network that we are currently 
> building called the PlanetLab Backbone (PLB) links all of these 
> devices together into a single fully connected mesh. The TUN/TAP 
> interface is generic, however; any slice may send and receive its own 
> packets via /dev/net/tun, and implement their own overlay networks 
> within their own slices. More details about PLB are forthcoming.
 From what I understand, given the details here, the solution
appears very complex and not scalable (full mesh) to me.

Wouldnt it be simpler to give permissions to individual vservers
to create tunnel interfaces? Alternatively PL can provide an api for
the application talk to a tunnel daemon to create the tunnels and
update the vserver configuration file with the additional device
created?

-Venkata
USC/ISI
>
> --Mark
> PlanetLab Operations
>
> _______________________________________________
> Users mailing list: Users at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/users