Dual Radios: Was Re: Virtual WiFi on Linux?

[Jim Thompson]

On Oct 19, 2005, at 11:49 AM, Kelly Hogan wrote:

> I guess my bad.
>
> We don't run our Dual Radio APs in dual master mode.  That would  
> assume
> that the backhaul would use some WDS architecture, which is extremely
> inefficient.  We run one in master, the other in managed client mode.
> The second client radio leaches onto the any of the other mesh nodes
> Master AP, as a client, excluding its sibling AP radio.   We then do
> routing magic.
>
> Our tests, with the unit running with the client backhaul radio
> connected, show a 5% decrease in throughput vs the same unit  
> without the
> backhaul radio running, and getting its backhaul from a hardwired
> interface.  If the two radios running together were the problem, seems
> we would see something different.

How do you know that the signal level from the "other end" isn't high  
enough to have set CCA
on both radios?

Are you running to a source that can supply data fast enough that  
multiple STAs associated to the AP
are attempting to use the network?

Are you attempting to maximize throughput *and* range, or are you  
trading off one for another?

You've proved nothing here.

And the issue isn't AP .vs STA .vs WDS.

>
> So if the problem is related to two APs in the same box in master  
> mode,
> then so be it.  I can't comment on that.  But two radios work fine  
> with
> little effect in performance or distance of clients to the master AP
> when configured as our mesh system.

I think you'll find, that with a single radio running software which  
can act as both client and AP, you'll have
approximately the same throughput as you do now with two radios.

>
> With hundreds of outdoor, long distance hotspots (campgrounds) around
> the country, I think we would be seeing something entirely  
> different in
> our support calls if it didn't work.


Great, I'll buy one of your units, and install it in my campground.    
I'll connect my "campground" to a mythical 100Mbps Ethernet fiber  
pipe (with some router in-between) with a video server on the other  
side of the Internet pipe.   I'll put an ordinary 54Mbs AP in front  
of the router that has the big Internet pipe.

Then I'll locate the campers "at range" for a given modulation  
rate.   That is, I'll arrange them in a circle around the tower at a  
range where they can maintain a 54Mbps connection, but I want that  
circle to be as big as possible.

I'll configure your unit to associate with the "ordinary 54Mbps AP",  
and then move your unit (and the circle of users) to a range where  
your unit can also just hold a 54Mbps link.

Then I'll tell the people in the circle to download movies at will.

If I can't get 20Mbps of aggregate throughput to the circle of people  
madly downloading movies, can I sue your company for false advertising?

If I can show a > 5% degredation in aggregate throughput to the  
circle of madly downloading campers when compared to a similar setup  
where the "backhaul radio" is connected to the 100Mbps Ethernet via  
100BaseFX, but the CIR of the IP connection is turned down to 45Mbps  
(or I install a DS3 rather than the mythical 100Mbps Ethernet to the  
campground), can I sue your company for false advertising?  Can I  
call your support line and claim your product is defective and insist  
that you made claims that aren't supported by your product's abilities?

Whats probably true is that the bandwidth into your campground  
"hotspot"s is helping mask the issue.

And what happened to showing me how you stay "FCC legal" and get a 4  
channel separation while running 3 channels?

I think you might want to google for "adjacent channel interference"  
and read some.

jim

> KH
>
> On Wed, 2005-10-19 at 15:11, Jim Thompson wrote:
>
>> On Oct 19, 2005, at 10:54 AM, Kelly Hogan wrote:
>>
>>
>>> LOL,  Come on Jim....
>>>
>>>
>>>> This is why, sans co-ordination (or a lot of filtering), two radios
>>>> (even on ch1 and ch11) in the same box don't work.
>>>>
>>>>
>>>>
>>> Someone tell our running 200+ dual radio AP Mesh Repeaters that they
>>> can't work..
>>>
>>> This is simply not the case.  We have lots units in the field,  
>>> running
>>> on Soekris 4511 & 4526, dual radios, AP Master, and Backhaul client,
>>> hostap code, and plenty of throughput.  One box, 3-4 channel
>>> separation,
>>> and smoken performance.  Our in field tests show less than 5%
>>> degradation in overall performance when adding nodes up to 16  
>>> nodes in
>>> the mesh.
>>>
>>> Jim, IT CAN BE DONE!  Lets be real!  Maybe on paper it causes  
>>> problems
>>> but our paperless engineers took the lets build it and see
>>> approach.  It
>>> works.
>>>
>>
>> Sigh.
>>
>> Please perform this simple test.
>>
>> Take one of your "dual radio" APs.
>>
>> Put both radios in 2.4GHz, one on channel 1, the other on ch11 (since
>> in the US, at least, you can't run above channel 11.)
>>
>> Turn one radio off.
>>
>> Take a client, and go to the limit of coverage.  Establish a high-
>> flow test (ttcp or similar), and note the throughput.
>>
>> Now, turn on the second radio in the AP, and take a second client and
>> perform the same 'throughput' test.
>>
>> Note that the combined 'throughput' of both clients is *less* than
>> they would have had were they on the same channel (and radio).  Go
>> ahead and test it.
>>
>> Now explain "IT CAN BE DONE" to me again and quit lecturing me that
>> the relatively simple math on paper doesn't apply to the real world
>> because you've "seen it work".   Many people have seen "perpetual
>> motion machines" and "200mpg carburetors" work too, yet both are
>> simple to disprove using only elementary mathematics.
>>
>> Every time I've challenged someone with this test they've gone  
>> silent.
>>
>> Yes, if you get the received signal strength at the AP high enough,
>> it will appear to work.  (This is why I sent you to the edge of
>> coverage (pick your modulation speed to define the limit of
>> coverage.)     If you choose this path, all you've really done is
>> reduced range for no more throughput.
>>
>> If you get *close enough* then the signal arriving from a remote STA
>> on ch1 will be high enough that the radio (in the AP) on ch11 will
>> *SET CCA* and no longer transmit for the duration of the incoming
>> packet (plus a DIFS period.)
>>
>> If you separate the APs enough, (in space and frequency) then the
>> problem (largely) goes away.
>> If you install a bunch of filtering at the AP (think 40dB+ channel
>> filters for ch1 and ch11) you can make it work.
>> If you co-ordinate the operation of the radios in the AP, you can
>> make it work.
>>
>> Also, please explain how you got a "4 channel separation" between the
>> two radios and stayed legal.
>>
>> I am being (very) real.
>>
>> jim
>>
>>
>>
>>
>>
>