[OLSR-users] Re: Networking help on Metrix Mesh

[Benjamin Watson]

I haven't run exhaustive bandwidth performance tests yet, as I'm just
trying to get it all to "work" for now.

With L1 connected to R1 which is connected directly to the Internet,
L1's web surfing performance seems good (similar to connecting to a
traditional off the shelf AP router).

However, I found out that problem #1 I described previously is
inaccurate.  It turns out that L2 can pull up the default web page
hosted on L1 via the mesh!  I must've been impatient when the page
didn't load up right away.  But, the performance is horrible!  It took
well over 30 seconds to get a simple HTML "It works" page to load.
Occasionally, the connection times out.

FTP, on the other hand, "seems" to perform somewhat better.

I've read several studies on implementing single radio meshes, where
the radio serves both the end user and participates in traffic
forwarding to form an ad-hoc "backbone".  Under these scenarios, I've
seen the trend stated that bandwidth will drop on the order of 1/n^2,
where n is the number of hops.  The primary reason for this is that
the radio serving the end user cannot listen to the end user as it
transmits data to an upstream node, which is the inherent nature of
wireless communications.

James, it seems that the performance issue you describe is a direct
result of the equation above.  Check out www.meshdynamics.com.  The
site's author goes into more detailed discussions as to why this
happens.  Ideally, I'd like to do what the author proposes and use 3
radio mesh routers, which "should" eliminate the performance issue.

However, my 2 radio nodes "should" also mitigate this.  Radio A is
using 802.11a with OSLR to form the "backbone" network.  Radio B is
using 802.11b functioning as an AP to serve end users.  Therefore, the
radios should not interfere with one another and be able to
send/receive simultaneously.  The only performance penalty here, in
theory, is the routing going on in the kernel, which shouldn't be too
bad.

I was also concerned about possible interference between L1 and L2 and
the APs between R1 and R2, which is why I have the 802.11b radio on R1
set to channel 1 and the 802.11b radio on R2 set to the other end of
the spectrum (channel 11).  So link1 and link3 as I described in my
"ascii art" diagram earlier shouldn't interfere.  So why is my
performance so horrible for HTTP?

James, you are correct that R1 gets its eth0 IP via DHCP.
Furthermore, R2 sets R1 as its default gateway once OLSR has settled
down and updated all of the tables, but DNS still isn't working on R2
or L2 (neither know what www.google.com is).  This is why I think I
need to learn how to properly configure the OLSR nameservice plugin
and/or DNSMASQ.

Ben (nice name by the way), both of my radios on each Mesh node are
Atheros cards.  However, I don't have any antennas just yet (which
could be a reason for performance issues, however everything is
located in the same room at this point no more than 3 feet from one
another).  Furthermore, the weatherproof boxes would have to be
modified to accomodate 2 extra antennas as each radio currently is
only set up to use a single antenna.  Say, that might be the problem.
Perhaps I should explicity disable antenna diversity.  I've also been
playing around with RTS and Fragment thresholds to see if that helps.

Any other tips or tricks are welcome.  Thanks for the replies thus
far, gets my brain thinking about something else to try.  I'll let you
know how things go.

On 11/15/06, Benjamin Henrion <(spam-protected)> wrote:
> On 11/15/06, James Wilson <(spam-protected)> wrote:
> > Hey Ben,
> >
> > I was wondering what kind of bandwidth you're getting with the dual
> > radio (per router) configuration ?  I've implemented a small 5 node mesh
>
> If you use 802.11a with dual radios per node, you can expect 25 MBPS
> if all the radios are at the maximum bitrate.
>
> I am thinking about this idea:
>
> 2 atheros cards per node with card1 = TX power at the max + RX
> sensitivity at the minimum and card2 = TX power at the min + Rx
> sensitivity at the max
>
> Tighted together with bonding0, with OLSR listening on Bond0.
>
> Having this setup would split which card transmits on a fixed channel,
> and which card receives on another fixed channel.
>
> > each with single radios, forcing traffic to go through multiple nodes,
> > the throughput is horrible !   I get 6 or 7 mbps going through 1 hop,
> > anything more it drops to 300kbps   :(  How much better does a dual
> > radio setup function  ?
>
> Most academic papers says that you can achieve better perfs without
> reusing the same channel.
>
> The problem is how to adapt OLSR to this (maybe there is a way to do
> it a lower OSI level).
>
> If someone has the hardware to make a test (2 atheros card per node),
> that would be useful to know how OLSR behaves.
>
> --
> Benjamin Henrion <bhenrion at ffii.org>
> FFII Brussels - +32-484-566109 - +32-2-4148403
>
> _______________________________________________
> olsr-users mailing list
> (spam-protected)
> https://www.olsr.org/mailman/listinfo/olsr-users
>