[Olsr-users] lxc network emulation

[Markus Kittenberger]

On Sat, Oct 22, 2011 at 1:00 PM, <(spam-protected)> wrote:

> Hello again Markus,
> I would like details about the topology you had:
>
> Here, you seem to speak of a network where olsr are isolated,

no they aren`t isolated!
(it wouldn`t make much sense to run hundreds or thousands of isolated
routing daemons, wouldn`t it?)

have their own network stack, means, they do not borrow the lxc host ones,..
(which is also possible with lxc, but does not make much sense when
simulating networks,..)

i.e. they have there own interfaces, ip adresses, and there own routing
tables,..
(they also forward traffic, and are pingable itself)

but they do not have there own "system" running,..
(instead of exectuing init of the rootfs, lxc configures all network aspects
itself, and directly starts olsrd)

but having an rootfs with an a serial console runing costs less than 1MB
RAM, but ~ one second additional startup time, but afterwards cpu load is
the same.

(which means ~ 90 minutes to wait for 2000 instances to start, while without
rootfs its doable in ~ 10 minutes )


> I really
> need precise figures to be sure we are comparing similar things, about the
> case of the 2000 lxc running olsr:
>
> Could you reach each node with each other node with a ping ?
>
sure,..

>
> And also: how many hops where necessary for the extreme nodes to reach
> each-other?
>
hmm average was ~ 7 hops.

but topology (except maybe extreme ones) do not matter much regarding load,
its (beside the number of isntances) the number of (olsrd) links that
affects the load (and memory consumption) of the running routing daemons.

>
> In my case all pings where good and reactive and the biggest number of
> hops was around 20.
>
> but next time i can try a large square or cube setup, to have a easier
compareable topo.

Markus
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.olsr.org/pipermail/olsr-users/attachments/20111022/b7614925/attachment.html>