[olsr-dev] A new idea for 0.5

Andreas T√łnnesen (spam-protected)
Thu Jan 5 14:35:36 CET 2006


Hi all,

Ok, since people are already replying to my (rather boring ;-) ) libnet
mail with new ideas, I figure I might come forth with one none-code
related new idea I have for the 0.5 series.

Let me start out by saying IANARE(I Am Not A Radio Expert), but I hope
this mail makes sense.
As of now we use the HELLO messages to calculate linuk quality. There are
som problems with this AFAIK. HELLOS are broadcasted/multicasted(IP) which
causes these packets to easily be lost on links with high load since there
is no ACK/retransmission of these frames. Also in 802.11b broadcasts are
sent on 2Mbit which may cause these signals to reach further than what
unicasted signals do, this might cause the link to "seem better than it
is".

What I suggest is that we add unicasted UDP heartbeat messages. In our
relativly non-mobile networks we do not need very low HELLO intervals. (To
be able to use low HELLO intervals without making link detection sluggish
we might consider generating HELLOS triggered by some sort of requests
from nodes joining the network). I suggest that we announce the interval
of the heartbeat messages in the message itself(and perhaps the HELLO as
well) and the reverse LQ and that this is the only payload in the message
for now. We might consider generating various sized heartbeats for more
thorough link detection later on. An example configuration would be:

HELLO interval: 10 secs
HELLO validity: 60 secs
HEARTBEAT int: 2 secs

A node would then every 2 seconds, unicast a HEARTBEAT message to all its
links to all neighbors(remember there can be multiple links to a
neighbor).
Neighbors will now calculate their LQs based on this traffic and not on
the broadcasted HELLOs. Yes, this will cause some overhead, but I think we
can live with that?
What we get in return is less packetloss of HB vs. HELLOs on busy links
giving us a more realistic LQ. The HELLOS will still be used for neighbor
detection and link establishment but they will have no real active part in
"link maintainance".

(Hmm... this was all written a bit to fast, but I hope it was understandable)
Does this sound like something worth giving a shot?

- Andreas





More information about the Olsr-dev mailing list