[Battlemesh] recommendations for 40Km Link

Dave Taht dave.taht at gmail.com
Tue Mar 25 02:05:58 CET 2014


On Mon, Mar 24, 2014 at 2:58 PM, Nicolás Echániz
<nicoechaniz at altermundi.net> wrote:
> El 24/03/14 00:12, Dave Taht escribió:
>>
>> On Mar 22, 2014 9:25 PM, "Nicolás Echániz" <nicoechaniz at altermundi.net
>> <mailto:nicoechaniz at altermundi.net>> wrote:
>>>
>>> El 04/07/13 05:00, Joseph Bonicioli escribió:
>>> > Links of that distance are very common in our network (AWMN). Attika in
>>> > Greece is not flat resulting to many links that are over 10 20 or
>> even 50Km.
>>> > Moreover since we got access to TV channel installations we have a
>> lot of
>>> > long distance links to other areas of Greece and with other wireless
>>> > communities . It will be pointless to point some out. They are many
>> and I
>>> > own a couple of them.
>>> >
>>> > Do not over analyze. Maths never fail you (FSL calculations)  and rocket
>>> > dishes are fine for 40Km. With 34dbi dishes at both ends you will
>> get -71db
>>> > signal with no Tx power at all. If you add 5-10db Tx power you will
>> be well
>>> > in the good performance zone of -61 to -66.
>>> > If you are in a noiseless environment you will be thrilled to find
>> that the
>>> > performance will be "just" as good as in your test room.
>>> > You can also find cheaper alternatives. Get yourself a couple of
>> good solid
>>> > reliable 120cm offset dishes, a couple of feeders that we can
>> provide you if
>>> > you need to and you are ready.
>>> >
>>> > On the other hand don't consider anything less than the biggest gain
>>> > (biggest dishes) and most reliable hardware you can get. You will
>> have to
>>> > compensate for bad weather in the winter making the link harder due to
>>> > conditions, harsh environment may punish hard you equipment and of
>> course
>>> > unreliable hardware/software will make you travel a lot. The difference
>>> > between some extra cash for something a bit more reliable and traveling
>>> > expenses is big.
>>> >
>>> > And don't worry about details like alignment. Those things nowadays
>> if you
>>> > have not broken something in your equipment are a breeze with the
>> tools that
>>> > you get.
>>> > Plan for other details if you are going somewhere far or on a mountain.
>>> > Cables, extra tools, insulation material, things that make your life
>> and the
>>> > experience pleasurable, communication equipment, ropes, toolboxes,
>> people :)
>>> >
>>> > Go play, take some pictures and let us know the results! Good luck!!!!
>>>
>>>
>>> We finally got around to finishing our first long distance (50Km) link
>>> and it's working great.
>>> Thanks joseph for the hints and optimism :)
>>>
>>> We published a small blog post (in spanish and english):
>>> http://blog.altermundi.net/article/completamos-el-enlace-de-50km/
>>>
>>> with some details on link conditions, quality and photos.
>>
>> I am impressed. Was the link really 50 Mbit in both directions
>> simultaneously? Are you using airos or OpenWrt?
>
> We plan to use OpenWrt everywhere when ath9k gets more stable, so far we
> have had different issues with every release we tested for the past year
> or so.

tell me about it.

> So here's what AirOS "speed test" is reporting:
> http://blog.altermundi.net/media/uploads/captura_estado-enlace_zoom.png

and I don't trust whatever it is doing without packet captures.

> running it repeatedly gives similar results every time.
>
> Check the blog post for other images, like terrain profile, etc.
>
> http://blog.altermundi.net/article/completamos-el-enlace-de-50km/
>
>
>> Got Bufferbloat? What does a rrul test look like? (See netperf-wrappers)
>
> We'll have to wait till we can upgrade to OpenWrt

No, the behavior of airos is interesting in itself. I haven't tested
it in years and certainly not at this range.

The rrul test runs on any linux or macosx box. It requires netperf 2.6
compiled with --enable-demo,
fping, and python-matplotlib for plots. Topology:

laptop (via ethernet preferably) -> wireless router -> wireless router
-> server somewhere.

you run netperf's "netserver" on the server (no other tools required),
and something like this on the laptop:

netperf-wrapper -l 60 --disable-log -p all_scaled -o mytest.svg -H
server -t 'test title' rrul_be # or rrul

There is a new -gui option as well.

Regrettably I have no netservers in argentina (:() so you'd have to
convince someone
on a nearby hop in the university to run one for a test. server only
needs netserver running.

This is an example of the results I got with sqm on a cable system

http://snapon.lab.bufferbloat.net/~cero2/jimreisert/results.html

Wifi tends to be much uglier, and degrades terribly in rain.


netperf generally needs to be built from svn head

svn co http://www.netperf.org/svn/netperf2/trunk/ netperf; cd netperf;
./configure --enable-demo; make; sudo make install


on the client:

The github repo for netperf-wrapper is here

https://github.com/tohojo/netperf-wrapper.git


apt-get install python-matplotlib on the client # debian derived systems
# I have no idea what to do on other systems, there is some doc in
github about netperf-wrapper on arch.


>
>
>> How well does it work in rain?
>
> It hasn't rained since we set this up. I will report when it does.


>
>
> Cheers,
> Nico
>
>
> _______________________________________________
> Battlemesh mailing list
> Battlemesh at ml.ninux.org
> http://ml.ninux.org/mailman/listinfo/battlemesh



-- 
Dave Täht

Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html



More information about the Battlemesh mailing list