p2psp team mailing list archive

[Max Mertens <max.mail@xxxxxxxxxx>]

Hi Vicente,

On 08/17/2015 09:57 AM, Vicente Gonzalez wrote:
> This is the result for the test of the Java Applet you have indicated
> before [2], for the University of Almería:
>
> *******************************************************
> ...
> *******************************************************
>
> I think that if your developments can manage this NAT, you can hope to
> handle any other kind of NAT :-)
>
> As you mentioned before, maybe you can use also the NAT of your mobile
> device (I'll try to check also mine).
Thanks for that information! The network of my university shows similar
results, but with a UDP timeout of 16 instead of 0 seconds. So in the
tests this was good to use for splitter and monitor.
I tested the mobile internet, and apparently it implements a cone NAT
(probably a PRCN) so it has the same source port for every destination
endpoint.
As the connection was a bit slow, I had to increase the timeouts. With
the splitter and 2 monitor peers on the university server, one peer via
DSL and one peer via mobile internet, the NAT traversal between the
peers worked. When adding another peer in a virtual machine behind a
SYMRP NAT (unpredictable source port) to the team, the other peers could
not connect to it, as to be expected.
The free VPN services that I tried (and that were usable for tunneling
the hole networking) had symmetric NATs with random source ports, so a
peer could only connect to splitter, monitor peers and peers behind a
FCN or RCN NAT.
Do you suggest other testing combinations or methods?


On 08/17/2015 10:48 AM, Vicente Gonzalez wrote:
>
>     [2] https://asciinema.org/a/1wxkq1mf4oseiex1hzfn0gkst
>
>
> Hello Max,
>
> good work! Only for this example [2], could you send to me a timeline
> with the messages between peers, splitter, monitor peers, etc. in
> order to understand it better (and remind to me a lot of forgotten
> things :-)?
Thanks! Yes, this is the next thing I will do, adding sequence diagrams
with explanations to the documentation, about the NAT traversal and the
retry process. I will send you a link as soon as it is uploaded.

> Well,  this project has been a difficult one (it has a lot of research
> work and possible results). Therefore, the documentation is as
> important as (or even more than) the code. We (I and the rest of
> mentors of P2PSP) also think that your work could be published in a
> conference (such as the International Conference on Computer
> Communication and Networks <http://icccn.org/icccn15/>) or even a
> journal  (an example could be: Computer Networks
> <http://www.sciencedirect.com/science/journal/13891286>), and we would
> delight if you collaborate with us in this task. What do you think (of
> course, this will not affect to our work/evaluation/results for the GSoC)?
It would be great if it would be published in some way, as other P2P
software could also benefit from the NAT traversal methods; I would like
to collaborate with you in this.
Last week I also thought about maybe putting the implemented methods
into a portable NAT traversal C library, to embed in P2P applications.

The updated task list can be found here [1]; the only coding task left
for this week is the determination of the currently allocated source
port for incorporated peer behind a SYMSP NAT, which is needed if a new
peer arrives a long time after the last peer has arrived (so there could
be many source port skips).

In the last days I added a script to setup a network with NATs in Linux
namespaces, which works much smoother than with VMs; also the test
script now supports parallel test runs. During testing, I created
another script that runs a team and filters/colorizes the output (as you
saw in the terminal recording); I would add that to the tools/
subdirectory, if you could need this.

Thanks,
Max


[1]
https://github.com/jellysheep/p2psp/wiki/GSoC-2015:-NAT-traversal-using-UDP-hole-punching---Timeline