[Server-devel] [XSCE] Fwd: [support-gang] Number of XO-1 per AP on XSCE - one data point

[James Cameron]

Good question.

The easiest way to infer utilisation is to use the channel to transfer
data, and measure how much data can be moved in a time period, which
gives you a rate, which is then compared against theory.  wget is
enough for this, but with large files slows as they have to write to
disk.

An optimisation is to avoid using disk as part of the test, which
means using iperf instead of wget.

The complex way involves capturing packets and measuring the time
spent receiving them; tcpdump, wireshark.

Remember that utilisation is affected by all uses of the frequency,
not just the equipment you are testing.  I've always been able to get
great data rates on a farm, but take the equipment into a town and the
rate is much lower.

On Mon, Apr 07, 2014 at 07:17:43PM -0400, George Hunt wrote:
> That would satisfy my curiosity. 
> 
> Is there an easy way to sniff a channel to infer it's utilization? Maybe look
> at collisions/retransmissions/, or error rates. or . . .
> 
> Seems like 3 AP's might give us the same information with a lot less work.
> 
> 
> On Mon, Apr 7, 2014 at 7:11 PM, James Cameron <quozl at laptop.org> wrote:
> 
>     It would be interesting to see the same test with XSCE, ejabberd, and
>     three APs on separate channels.  This would reduce the processing
>     burden on the AP CPUs, and reduce the air time requirement.
> 
>     On Mon, Apr 07, 2014 at 06:55:55PM -0400, George Hunt wrote:
>     > This number resonates with me.  Kevin Gordon, in Toronto, was interested
>     in
>     > XSCE primarily because it would offer ejabberd, without which the
>     factorial N
>     > conversations taken two at a time would fill the air waves at N=13.
>     >
>     > I'm responding this way because of your statement that the clients were
>     > "non-XOs". My guess is that no AP will be able to handle more than 13
>     without
>     > ejabberd (and the registration process between the XO and the server
>     which
>     > enables it).
>     >
>     >
>     > On Mon, Apr 7, 2014 at 12:14 PM, Adam Holt <holt at laptop.org> wrote:
>     >
>     >     FYI!
>     >
>     >
>     >     From: Nathan C. Riddle <nathanr333 at charter.net>
>     >     Date: Mon, Apr 7, 2014 at 12:05 PM
>     >     Subject: [support-gang] Number of XO-1 per AP on XSCE - one data
>     point
>     >     To: Gang <support-gang at laptop.org>
>     >
>     >
>     >     The maximum usable number of XO-1's (12.1.0) appears to be 14 using
>     as AP
>     >     TP-Link MR3020 on XSCE 0.4 RC1 on XO-1.5  with about 10 non-XO's
>     showing in
>     >     neighborhood. Number 15 is unreliable at connecting. With 14, simple
>     >     connections to MOODLE appear reliable.   XSCE 0.4 was used since it
>     was the
>     >     only one available last September as school started.
>     >
>     >     Testing DKMS libertas.ko file provided by quozl.org (http://
>     dev.laptop.org/
>     >     ~quozl/12757/dkms     with chmod 744 to file) to establish baseline
>     for
>     >     Mesh Potato-2 Basic AP replacement of MR3020. Presence of XO-1 with
>     no
>     >     modification appears to have no effect  on connections by modified
>     XO-1's
>     >     (as contrasted to XO-1's with old .ko file).
>     >
>     >     Hoping MP-2 will push this past 25.
>     >
>     >     Supplying this one data point since I had previously ask this
>     question
>     >     here.
>     >
>     >     Nathan Riddle
>     >
>     >
>     >     --
>     >     Unsung Heroes of OLPC, interviewed live @ http://unleashkids.org !
>     >
>     >
> 
>     --
>     James Cameron
>     http://quozl.linux.org.au/
> 
> 

-- 
James Cameron
http://quozl.linux.org.au/