[CUWiN-Dev] tdv - interesting behavior

[Matthew Isaacs]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

As part of some work on tdv today, we would up with multiple gateways to
a cloud.  It worked fine for a while; quite nicely, actually.  Then, at
some point, things stopped working.  Some investigation reveals that the
likely culprit is hsls taking all the cpu.  When things were working,
load averages are under 1.0; when things broke, they had climbed to over
1.5 and were using 100% of the cpu.  I have no idea why hsls is taking
some much cpu.  The routes look fine.  The only thing I found that
looked odd were the etx metric associated with each route.  I'm assuming
that etx is like most metrics in that a lower cost means a better route.
 According to the linkstates table, however, the cost associated with
the ethernet on the gateways is set to the maximum cost (65536).
Restarting hsls does not seem to affect it.  tcpdump shows a large
amount of traffic on the wireless being generated by hsls; it appears
that the nodes are both able to send and receive.  My guess is that hsls
is simply using the radio too much to allow other traffic to pass.

I know we're working on gateway selection, and that has its merits.
However, I was thinking and such a mechanism isn't required in
traditional routing; the routing protocol and an appropriate metric deal
with that.  Additionally, this was working for 5 hours today before
things started causing problems.

For the time being I'm just going to group things into clouds, but I'd
be interested in people's thoughts on this.

- --Matt
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFF8fjNFTAfZok65zkRAqD1AJ9Q5Qt3sdQCzNsBds0m5yWzN8VItwCeKW4X
o8+g5EN3b6vjd5MIx0Cpe+g=
=ld7b
-----END PGP SIGNATURE-----