Hi Keith,
Thank you so much for your time and help!
I tried this approach with both switches having an interface directly
connected to the host, and it did work perfectly.
However, again, if I only let vigil1 has an interface, run secchan on
vigil1 normally and run secchan on vigil2 with [CONTROLLER] omitted (let
it discover the controller), the same problem (DHCP packet data too
short to parse header) occurs. Another interesting thing is that if I
launch the "switch" application instead of "pyswitch", the dhcp.py in
/nox/src/nox/lib/packet does not get called. But to me controller
discovery seems necessary, since in a larger network not every switch
could be directly connected to the controller.
Thanks again,
-Zheng
Keith Amidon wrote:
I provided the example "2 host 2 switch" configuration awhile ago.
I'm not sure I can reproduce your problem in my environment as my VMs
are probably not running the same OS, etc. as yours.
For what it is worth, with that configuration in my VM setup I find
that at startup the vigil1 VM gets an IP address on the interface
connected to the "host" computer via DHCP, but the vigil2 VM does
not. I've not investigated deeply why this is, but suspect it has to
do with the default way the OS in my VMs attempt to bring up
interfaces. It's also been suggested this may be a VDE issue, but I
haven't had time to investigate further.
In any case, I have to manually bring up the interface to the "host"
on vigil2. The only thing that is tricky about this is identifying
the correct interface since vigil2 has multiple interfaces. To
determine the correct interface, look at the output printed when the
VMs are started. I've selected the two important lines for my situation:
...
slirpvde -daemon -socket /home/keith/vm/vde/ctls2 -dhcp
...
screen -c /home/keith/vm/vigil2.screenrc -dm -S vigil2 -e ^Oo vdeq kvm --mod
700 -m 32 \
-hda hda.dsk -kernel kernel.bin -append 'root=/dev/hda1 console=ttyS0
apm=power-off noapic ' -nographic \
-net nic,model=rtl8139,vlan=1,macaddr=50:54:00:00:00:04 -net
vde,vlan=1,sock=/home/keith/vm/vde/ctls2 \
-net nic,model=rtl8139,vlan=2,macaddr=50:54:00:00:00:05 -net
vde,vlan=2,sock=/home/keith/vm/vde/ctls3 \
-net nic,model=rtl8139,vlan=4,macaddr=50:54:00:00:00:06 -net
vde,vlan=4,sock=/home/keith/vm/vde/ctls5 \
-snapshot
...
In the first line shown, slirpvde was supposed to run DHCP on
/home/keith/vm/vde/ctls2. This was what should have given vigil2 an
IP by DHCP but didn't work. In the second line, notice that
/home/keith/vm/vde/ctls2 is associated with the nic with macaddr
50:54:00:00:00:04. With this information, you can identify the
interface on vigil2 by looking at the output of 'ifconfig -a' run
within that VM:
...
eth1 Link encap:Ethernet HWaddr 50:54:00:00:00:04
...
Since the HWaddr of eth1 matches, that must be the one connected to
the "host". The network that is created for the connections to the
"host" is 10.0.2.0 with the host on 10.0.2.2. I manually give an
address to the eth1 interface in the vigil2 VM with the command:
ifconfig eth1 10.0.2.16 netmask 255.255.255.0
After this, I can ping the host computer, and I can connect secchan to
a controller running on the host using the 10.0.2.2 address. As I
said, I'm not sure whether you are experiencing the same issues as I
am. Hopefully this account at least gives you some hints for where to
look to resolve your issues.
--- Keith
On Wed, 2009-04-08 at 14:43 -0500, Zheng Cai wrote:
Hi,
I have tried both "ifconfig of0 10.0.0.5" and "secchan nl:0
tcp:10.0.2.2:975" on vigil2 too, but the behavior is still the same.
I.e., after 45 seconds, the switch tables start to contain active flows
and forward end host traffic, but still with error messages "Could not
connect to controller, failing open".
However, I found out that, on the nox_core side, warning messages "(dhcp
parse) warning DHCP packet data too short to parse header: data len 86"
keep popping up. Was this the reason for such behavior? If so, any
suggestions on how to solve this issue?
(BTW I used the latest NOX version, downloaded the kernel.bin and
openflow_mod.so and secchan from noxrepo.org. I used in-band control,
ran nox_core with the "pyswitch" application. I added the nl:0 dp, and
all eth interfaces to the dp except the one connecting the host, on both
vigil1 and vigil2)
Thanks a million,
_______________________________________________
nox-dev mailing list
[email protected]
http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
