On Thu, Mar 05, 2009 at 02:22:15PM +0100, P. Mazart wrote:
Am Donnerstag, 5. März 2009 schrieb Marek Lindner:
This paper just mentions the possibilities - not all flavors implement that feature. As far as I know bmx (batman-experimental) supports it though.
Now I'm completely stuck with BM's brands. :-)
Please read http://open-mesh.net/wiki/BranchesExplained to understand basic differences between the "brands". B.A.T.M.A.N.-adv has completely different ideas about what to route. It is not meant as drop-in replacement and not for any performance reasons.
Can you say something about overhead and performance of BM in kernelspace compared to the udp-userspace deamon? And devel-status of each of them?
You can only compare batman-adv-kernelland and batman-adv-userspace performance wise. Userspace has some performance problems and was meant to test the concept of routing ethernet frames instead of creating routing tables for ip routing. B.A.T.M.A.N. (aka batmand) has no kernel space implementation, but a supporting kernel module for it's gateway functionality. It should help the gateway node to reduce the performance bottle neck by not copying all data from kernelland to userspace.
If you would really compare batman-adv-kernelland and batmand then please ask again - I don't have any good numbers here, but maybe somebody else has.
If you really want to test batman-adv-kernelland then please use the last stable release from http://downloads.open-mesh.net/batman/stable/sources/batman-adv-kernelland/ and not the current version from svn. New experimental features were added there recently and it is not in a stabilising phase at the moment.
[…] The packets sent to the vis server are UDP packets […]
Ahh. Great. Thank's for the information.
If you want to check the amount of time between two vis packets: http://open-mesh.net/browser/trunk/batman/batman.c?annotate=blame&rev=12... It is around 10000ms at the moment in batmand, but as Marek said it depends on the implementation.
Best regards, Sven