[OLSR-users] Problems with some net clients
Tue Feb 7 15:41:39 CET 2006
This is somewhat orthogonal to OLSR, but since there are several folks with
large scale olsr deployments, hopefully someone can help.
I have a client buried deep in the mesh (i.e. 2-3 hops to get to an edge
router) that wants to use MSN Messenger. His problem is that Messenger will
only stay connected for limited periods of time (i.e. 3-5 minutes). His
disconnects seem to be timed to coincide with network architecture changes
(i.e. his route might shift from 192.168.7.44 -> 192.168.7.45 -> 192.168.7.1
(edge) to 192.168.7.44 ->192.168.7.40 -> 192.168.7.1 (edge) ) He doesn't
appear to be changing edge routers, but that shouldn't matter anyway, since
both HNA nodes shoot PTP to a final core router that NATs them on to the
internet from a single address. The bottom line is that, although the final
address that the MSN server sees should never change, the routes that those
packets take before the final NAT does change, and Messenger hates this.
Has anyone experienced anything like this before? All routers in the network
are WRT54G routers running olsrd 4.10.
BTW, I experimented with LQ versus Hysterisis, and it was the worst 7 days of
my networking life. Load averages on all routers went from .05 to 1.0+,
customers were constantly screaming at me about slow connections, poor
performance, and I wound up refunding about 1/2 my revenue for the month. I
was using "standard" settings with fisheye for a very static network (no
mobile nodes). I admire everyone that built and tested the LQ code, and I'm
jealous of those that use it in production, but my experience with it has
been absolutely horrible.
More information about the Olsr-users