[OLSR-users] [Fwd: [WLANnews] Two very interesting studies about TCP-performance in ad-hoc multihop networks]

[Sven Wagner]

hi
maybe, someone is interested ... ;-)

regards
	~cven

-------- Original Message --------
Subject: [WLANnews] Two very interesting studies about TCP-performance 
in ad-hoc multihop networks
Date: Thu, 30 Sep 2004 13:33:47 +0200
From: onelektra <(spam-protected)>
Reply-To: (spam-protected)
To: (spam-protected)

For those interested in the development of ad-hoc routing these two
studies could be very interesting. We were setting up a chain of
meshnodes in djursland to test the throughput across a multihop ad-hoc link.

The indian institute of science in bangalore and the computer science
department in los angeles did two researches on the same issue. The
results are quite encouraging.

They were doing laboratory tests and simulations. Here a short summary
of what I came across:

TCP has great performance problems in such a environment and the whole
network gets unstable with bigger TCP-Window sizes. But it stabilizes
very well with small window-sizes that TCP normally doesn't use. Also
the rts/cts-mechanism could be improved. Both institutes came up with
ideas of improving performance.

With increasing number of hops, the TCP throughput greatly reduces
initially and then stabilizes at 1/6 of the rate they had with a direct
peer-to-peer link. In theory you could have a chain of n > 7 nodes and
the result from end to end is about the same if each link is stable.In
theory, of course.

LA department of computer science also tested and simulated the
throughput in a grid and they initiated different data flows at the same
time. In a mesh grid with 200 randomly placed nodes they initiated 20
flows and had a throughput of 50 Kbit for each flow. The simulation was
done assuming the use of 2 Mbit links (old lucent/orinoco cards)

Well, 50 Kbit aint sound that much and LA didn't test with a
routing-protocoll which would add overhead - they were setting up static
links. 50 Kbit is a bit slower than a single channel ISDN-Line in
germany - but - this is what I always expected from a mesh:

A way to cover a big area with a minimum of power and effort. A network
that increases coverage with every computer going online. Of course it
is only a high speed network when traffic goes across a few nodes but I
would love to have a peoples-everywhere-network that offers 5 Kbytes.

Here the studies:

The Impact of Multihop Wireless Channel on TCP Throughput and Loss:
www.ieee-infocom.org/2003/papers/43_01.PDF

Performance Enhancement of TCP on Multihop Ad hoc Wireless Networks:
http://www.cs.concordia.ca/~sabee_an/adhoc.pdf

We should test the berlin-mesh with smaller window-sizes soon. Bangalore
  and LA had very good stability by reducing the window-size to 4.

cu elektra


_______________________________________________
WLANnews mailing list
(spam-protected)
http://freifunk.net/mailman/listinfo/wlannews