[CUWiN-Dev] Throughput

[Stephen Ronan]

Thanks all. Here is some more recent data from Sanjit (and btw, it seems 
they found a fair amount of additional bugs once they scaled up over 
thirty or forty nodes... if you're relying to some degree on their older 
code base, I wonder whether you might run into some of those same bugs 
at some point. I'm not myself expert in these matters by any means).

"Roughly speaking, the overall throughput will be halved at each hop 
because the radios are half-duplex (they can't transmit and receive at 
the same time). On the Roofnet, we've found things are a little worse 
than that, because hops along a route tend to interfere with each other.

"Here's a table showing the throughput from 44 of the Roofnet nodes to 
the three gateways:

     Hops    Nodes    Avg. TCP Throughput    Roundtrip Latency
       1      18          357.2KB/s                9.7ms
       2      10          112.0KB/s               17.5ms
       3       9           52.8KB/s               43.7ms
       4       7           47.3KB/s               43.0ms

"We're using 200mW 802.11b radios. If you work out the numbers, you'll 
notice that the results are significantly lower than the advertised 
throughputs for the radios. Accouting for protocol overhead, at 
11mbit/s, you'd expect a throughputs of 600KB/s for 1-hop, 300KB/s for 
2-hop and 200KB/s for 3-hop. The numbers we see are much worse than that 
because the most of the links show significant packet loss. We're also 
experimented with 54mbit 802.11g radios and found they actually do worse 
than the 11b radios."

 - Stephen Ronan

Sascha Meinrath wrote:

>>> Do you have similar experience in regard to that? About how many 
>>> nodes are in the largest current CUWiN installation? And what's the 
>>> most number of hops you see on a regular basis from any nodes?
>>
>>
>>
>> I lost count.  Sascha?  Three hops on a regular basis.  More when the 
>> testbed is more nearly complete.
>
>
>
> We know we're losing bandwidth with each hop, but haven't studied it 
> in any systematic way.  Currently we have about 20 nodes on our 
> network (though it's still not up and running again yet).  Most of our 
> nodes are 2-3 hops.
>
> Sorry for the brevity -- I'm just getting caught up on things after 
> being out of town.
>
> --Sascha
>