Hi Harald,
Holger has just refreshed my memory on a conversation we've had on running multiple NITB instances on the same box: Linux Network Namespaces vs. making NITB configurable.
Kindly remind me of the immediate plans, should we make osmo-nitb configurable to bind on only specific IP addresses, like, now?
IIUC that would be a VTY item for osmo-nitb and a parameter into libosmo-abis. Concerning the VTY, the port could be configurable, or the IP address, thinking of 127.0.0.42.
I guess relying on `ip netns` would be quicker for now, and I'd have slightly more time for IuCS.
Is there anyone else out there that would love to configure osmo-nitb to bind on specific IP addresses only?
Thanks, ~Neels
Hi Neels,
On Mon, Feb 22, 2016 at 11:39:06AM +0100, Neels Hofmeyr wrote:
Holger has just refreshed my memory on a conversation we've had on running multiple NITB instances on the same box: Linux Network Namespaces vs. making NITB configurable.
Kindly remind me of the immediate plans, should we make osmo-nitb configurable to bind on only specific IP addresses, like, now?
yes, I think that would be very useful.
IIUC that would be a VTY item for osmo-nitb and a parameter into libosmo-abis. Concerning the VTY, the port could be configurable, or the IP address, thinking of 127.0.0.42.
also don't forget the control interface.
Furthermore, the unix domain sockets for rf_control and for MNCC need to be configurable, too. Ah, I just see it already is by means of the '-r' command line argument.
In general I prefer things being configured via VTY paramters over command line arguments.
I guess relying on `ip netns` would be quicker for now, and I'd have slightly more time for IuCS.
I'm not sure this solves the general problem, and it is of course highly Linux-specific. For an all-IP NITB there currently is no strict Linux dependency.
I would say maing the addresses configurable seems reasonably straight-forward to do it right now. It shouldn't take much time, or am I missing something?
On Mon, Feb 22, 2016 at 12:11:27PM +0100, Harald Welte wrote:
I would say maing the addresses configurable seems reasonably straight-forward to do it right now. It shouldn't take much time, or am I missing something?
No, I think we've mentioned all items, thanks for the ctrl iface reminder.
The only counter argument is that I'm constantly doing things that don't take much time instead of moving ahead on IuCS :P
But indeed, it makes sense to do it now so that we can employ it in our test setup, saving reconfiguration time later.
I'll do that, then. (before hunting the hnbgw segfaults mentioned in another mail)
~Neels
-
Harald Welte -
Neels Hofmeyr