Two turrets

If both turrets are on the same physical axis then you would use positive X for one side and negative X for the other. Tool offsets make the details work.

If there are separate axes then us X for one slide and U for the other.

kostyadnepr wrote:

If I understand "The UV, WU and VW planes do not support arcs."

Good point, I had forgotten that.

Also, technically, you would be trying to make UZ plane arcs, and that whole plane is unsupported.
Incidentally, this is actually a limitation of G-code. To programme centre-format arcs you use the offsets IJK to define the centre position, and the letters LMN for the ABC equivalents and OPQ for UVW are used for other things.
In principle Radius-format arcs would be possible.

The solution would be to programme the arcs as a series of short straight moves. A lot of CAM systems actually do this anyway. Difficult to do by hand, though.

Actually, I don't see any reason that you couldn't set up the extra tool slide to be a Y axis. It isn't geometrically a Y axis, but the controller doesn't know that.

kostyadnepr wrote:

Now in fact, both of turrets have independent control on the X and Z (two "independent" pair of axes) and can work simultaneously. Each of the turrets has a NC program. And it seems to me it is necessary to use two LinuxCNC to recreate the behavior of the old controller.

Yes, that is probably the case. You can "fake" it in the G-code of a single controller, but not automatically.
What I mean is that if your current controller has the following (both starting at (0,0) for illustration though that would be expensive in practice.
G1 X1 100 F100
G1 X2 50 F100 / G1 X2 0 F100

ie, while X1 travels to 100 X2 travels to 50 and back again

You could "fake" that with a single controller that uses the Y-word to move the second axis:
G1 X 50 Y50 F100
G1 X 100 Y0

This isn't exactly equivalent, you don't have direct independent control of the feed speeds. You might be able to work round this with inverse-time mode and some clever preprocessing.

Is it possible to use the LinuxCNC as a client-server and to run the GUI on another computer?
In this way it will be possible to see differentiate interfaces of both channels (both controllers) on a single monitor.

This is something that the Machinekit project is much more flexible and capable at (largely because they like the Beagle Bone board, but that isn't powerful enough to run a GUI comfortably).

It would be very nice to be able to run two instances of the LinuxCNC motion controller in the same HAL pin space. But that is not currently possible AFAIK.