[Xen-users] XEN as virtual wireless enviroment

[xenml@xxxxxxxxxxxxxxxxx]

Dear all,

We want do develop a virtual wireless enviroment emulator at the
university of Düsseldorf (Germany) to test some new wireless protocols as
a substitute for the not well performing TCP.

The main problem is that the currently available wireless hardware has
implemented many protocols of the IEEE 802.11(a,b,g...) family hardcoded
in their hardware or in non opensource firmware.

But these protocols interfere with our protocols so we can't use any real
hardware to test the protocols in the wild.

At this point there comes a virtual network to mind which emulates
wireless behaviour like packetloss and the most important colition.

We already got a simple emulator from the university of Marburg (Germany)
which is based on Xen 2.0. This emulator uses a modified version of the
netshaper virtual network device and adds some functionality so set the
packet loss rate and some the bandwith limit.

But this emulator doesn't support collition which we need the most because
the new protocol should avoid packet collition.

So our first thought was to change the interface of the netshaper device
to also inform about collition.
At first we started some speed tests to see how fast we can inform a Xen
node that there was a (virtual) collition. For that we implemented a
simple UPD sender and receiver which sends UDP packets of a given size and
waits for the return and gives out the time.

Thats where I come to my question (sorry for the long introduction) to you:

The mean time the packet took to return was 3 milliseconds on a Pentium
Mobile with 1,6 Ghz and 1 GB RAM running 2 instances of Xen.
(OK, I forgott to mention that the Xen system was running in VMware 5.5)
What puzzled me most was the fact that the CPU load was 100%.

So my question to you:

Do you think that the VMware caused the performance problems?
When not, does Xen 3 perform better with network connections?
And finally: We would need a performance gain about 100 times, so about
0,03 milliseconds to return the packet from one xen node to the other and
back.
The bandwith isn't important, about 1Mbit would be fairly enough.

Do you think about any way of getting such good response times on a normal
hardware (1 CPU, Pentium 4 with about 2,5 - 3 Ghz) when using about 10 Xen
nodes?

Thanks alot in advance,

Greetings from Germany,

Matthias Jansen


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users