Hi all,

I'm in the process of putting together some routers which do load sharing over multiple channels (btw this is for mesh routing with the olsrd). At the moment i've got the multi radio multi channel part to work quiet well however i'm sharing the load by simply splitting half the routing table into one channel and the other half into another channel.

What i really want to do however is to make the system smarter by taking into account things like the signal to noise ratio of a channel and better yet the tx and rx buffers for the wireless interface (or anything that would give some sort of indication as to how busy a particular channel is ...maybe even the number of RTS and CTS received).

So what i wanted to ask is if anyone has any ideas about how i could get information like how full the tx and rx buffers are or RTS, CTS info or just any idea as to how we could implement this.

Cheers,
Victor

eheh... once you get those informations you are seeking, you have to manage them... this mean you need more cpu power to manage them. You could soon or late find out that you can't proceed like that unless you have a low-cost embedded platform that can realtime manage all those informations. At this time the platforms that can do that are too big or too expensive or difficult to buy or...
Olsr and its successor (Batman) are, among other things, an attempt to solve this issue.

The only advice I can give you is to deeply read their documentation and experience a bit their protocols before going over. Beacouse without that you risk to waste your time. I say this beacouse in the past years I saw many people/workgroups doing the same mistakes. And that's a pity.
As an example you can google around many attempts to bring l3 duties to layer 2... attempts that soon revelated to be a failure (OSI layer exists, among other things, to split networks complexity; if you mix l2 and l3, you get nothing more than a more complex system that require more resources to be managed).

Respectfully IMHO

ciao

Wallace

Hi Wallace

Thanks for the advice. However i'm afraid i've already invested too much time into this to back out now (I'm doing this as part of my honours thesis work). And i understand what u mean by needing more CPU power to manage the info as that is already one of the issues that is cropping up.

However i dont intend on mixing the l2 and l3 too much. Really just need a figure from somewhere to get a better idea as to the number of routes the routing table should forward in one direction and how many in the other. So it should only be a small additional load. Thankfully my goal isnt to make it perfect, but a proof of concept....some other poor kid who inherits this thesis topic will get the joy deciding whether or not its a dead end

I'm not saying that you must back out; I'm just saying that you should have a closer look to the past work made by communities already working on those stuff. Ex: measuring s/n or buffers doesn't help much to regulate congestion... I mean, if that is your goal, ETX is a good all-inclusive parameter... there are too many stocastic issues involved in wireless networking to go measuring them one-by-one... it's not only a practical problem of having/not-having enough cpu power... it's a totally wrong approach to something that simply CAN'T BE FULLY MANAGED. Ask yourself why the international working groups worked both on centralized AND distributed coordination functions...

You need to proceed fuzzy or you'll not have a proof for your thesis concept