Mesh testing on a cluster

[Javier Cardona]

Brian,

2009/3/16 Brian DeLacey <bdelacey at gmail.com>:
> We plan to try bridging meshes, using the MPP instructions at the end of the
> HowTo .

I believe that documentation was misleading (I've just clarified it in
the wiki).
The MPP functionality at this time only allows the mesh to bridge a
number of networks together, and not to bridge multiple mesh networks.
 In other words this is OK:

[ 802.3 network ] -- [MPP] -- [mesh] -- [MPP] -- [ 802.3 network]

and this is NOT ok:

[mesh] -- [MPP] -- [ 802.3 network] -- [MPP] -- [mesh]

For the latter to work, mesh portals would have to send path responses
on behalf of nodes outside the mesh, which they don't.
This functionality was written by YanBo, and in his patch you'll find
his description:

http://markmail.org/message/saansouwyr2fzurc?q=yanbo+mesh&page=1&refer=uf22u4zxr542ieca

Cheers,

Javier



> 2009/3/16 Francesco <francesco.cappuccio at gmail.com>
>>
>> Network manager was evil for me...on all machines involved in my tests I
>> wipe it away, installed wicd instead or configured it by hand.
>> Actually I managed to send on network some traffic generated with mgen, to
>> see how paths through nodes are chosen.Made some measurements. Pretty cool
>> except for that program is a little buggy..
>> Now I'm gonna see how the mesh behave with much more different shaped
>> traffic, video (tried a divx multicast and works like a charm :-) ), voip
>> and so on...
>>
>> Brian how many nodes were you able to interconnect?
>> Did you get to use brdges among different mesh or all nodes belong to the
>> same one?
>>
>> In case thing changed I ask again if anyone did succed into bridging the
>> mesh with Internet...any piece of code would be really appreciated :)
>>
>>
>>
>> 2009/3/16 Brian DeLacey <bdelacey at gmail.com>
>>>
>>> We made progress today getting a test mesh configuration established on a
>>> cluster. So far, we've only done a few basic tests, pinging nodes on the
>>> cluster, across the mesh but things are working well.
>>>
>>> We have a rack of clustered PCs and we've been able to ping those, via
>>> multiple hops, from laptops outside the cluster. The tricky part was how to
>>> test these remotely. We had to figure out a way to get around the usual
>>> first step, which is typically to "killall NetworkManager" so that we could
>>> continue to test via ssh remote logins. Thank you, Andrey, for providing the
>>> suggestion on how to handle this:
>>> Problem > Is there any way (or hope) to run mesh tests without
>>> shutting down NM? The cluster of computers we have access to has ongoing
>>> uptime demand.
>>>
>>> Solution > Tell NetworkManager to ignore the mesh interface, and the
>>> mesh-capable WiFi interface in general.  It's ok to have NM running as long
>>> as it leaves our mesh interface alone, otherwise it brings it up/down, tries
>>> to scan, etc.
>>> Now we move on to more advanced testing ...
>>> Brian
>>>
>>> _______________________________________________
>>> Devel mailing list
>>> Devel at lists.open80211s.org
>>> http://open80211s.com/mailman/listinfo/devel
>>>
>>
>>
>>
>> --
>> Francesco Cappuccio
>>
>>
>> _______________________________________________
>> Devel mailing list
>> Devel at lists.open80211s.org
>> http://open80211s.com/mailman/listinfo/devel
>>
>
>
> _______________________________________________
> Devel mailing list
> Devel at lists.open80211s.org
> http://open80211s.com/mailman/listinfo/devel
>
>



-- 
Javier Cardona
cozybit Inc.