ORBIT-USER: BSSID/IBSS madwifi idea?

[chris at orderonenetworks.com]

I think I have the ad-hoc BSSID problem solved.

I've adjusted the behaviour slightly to make it behave better with
non-tweaked nodes.

If there is a non-tweaked node that turns on first, (ie. better timestamp)
it will use that non-tweaked nodes' BSSID. If it gets there first, it will
make up its on tweaked/hashed value. This allows it to play better with
others, and yet if all nodes are running the tweaked version, you're
guaranteed they'll share BSSID's, since it either gets the BSSID from the
'head node' (that is a tweaked BSSID) or it makes up its own BSSID (which
is tweaked).

All that said, the code change is very simple. Working off the
baseline.ndz you need to open a file called:

/usr/src/madwifi/trunk/net80211/ieee80211_node.c

On line 427 you see code that looks like this (which you could probably
delete):

        if (ic->ic_flags & IEEE80211_F_DESBSSID)
            IEEE80211_ADDR_COPY(ni->ni_bssid, ic->ic_des_bssid);
        else
            ni->ni_bssid[0] |= 0x02;        /* local bit for IBSS */

Right after those lines add these lines:

        int i;
        for (i=0;i<IEEE80211_ADDR_LEN;i++) {
            ni->ni_bssid[i] = ic->ic_des_essid[i % ic->ic_des_esslen];
        }

Thats it.

And yes, the hash function is rather simple. However, I don't think it
matters much, since so long as the first byte is in the ascii range it
will not look like a MAC address on the ORBIT lab (which is where the
non-tweaked nodes get thier BSSID from if they don't find someone elses to
use).

to get all this running:

1. cd /usr/src/madwifi/trunk
2. make
3. cp /usr/src/madwifi/trunk/net80211/*.ko /lib/modules/2.6.12/net
4. modprobe -r ath_pci
5. modprobe ath_pci

and then do your standard iwconfig/ifconfig stuff.

I plan to run full tests tomorrow on the main grid and verify this, but I
feel pretty good about this solution - mainly because it alters the
original behavior in a way that should have no dependencies.

Chris

> Glad to hear we came up with the same idea.
>
> I really need to get this solved, so I hope I can get it implemented in
> the next few days.
>
> If you beat me to it though, don't hesitate to send me an email :)
>
> Chris
>
>> Nice timing. We were just in a discussion on this topic and the same
>> idea was proposed.
>>
>> My understanding is that the nodes are not seeing each other's beacons,
>> maybe in part because of some buggy implementation or as a result of
>> near simultaneous node-power-on in orbit.
>>
>> Anyways, for ORBIT like scenario, we agreed that a hash function was the
>> solution.
>>
>> So please go ahead with the idea.
>>
>> -Pandurang
>>
>> chris at orderonenetworks.com wrote:
>>> Hello All,
>>>
>>> I've spent a bunch of time reading madwifi source and google trying to
>>> come up with a decent solution to the BSSID problem where it jumps
>>> around
>>> on me.
>>>
>>> I think I've come up with a solution that I'm going to try and
>>> implement,
>>> but I'd like to see if anyone knows if it is a bad idea, or
>>> fundamentally
>>> flawed from the start.
>>>
>>> Basically in adhoc mode each node finds the oldest BSSID it can and
>>> uses
>>> that. Unfortunately this is doesn't work as advertised.
>>>
>>> The solution I'm thinking of implementing is simply hashing the ESSID
>>> (ie.
>>> iwconfig essid 'blah') into one of those XX:XX:XX:XX:XX:XX addresses
>>> and
>>> then when a node is in ad-hoc mode, I'll force the BSSID to that hash.
>>>
>>> Since each 'tweaked' node will use the same hash function they will all
>>> start off on the same BSSID. Any un-tweaked nodes will either follow
>>> along
>>> or do something else, but they should not be unduly affected.
>>>
>>> I like this solution since it appears that:
>>> 1. it will not muck with un-modified nodes
>>> 2. does not require any fiddling with the beaconing system, etc.
>>> 3. allows different ad-hoc networks with different ESSID's to co-exist.
>>>
>>> I'd appretiate any thoughts, comments, etc.
>>>
>>> Thanks!
>>> Chris
>>>
>>
>
>
>