OK. I will change that. Ah, 1990s nostalgia! ;-)
I considered that. But there are many use cases where you specify port manually, so this could be a feature. For example, you could manually assign ports in the range 5900-5999 for “front-end” VMs, with a convenient display number, but let “back-end VMs” be assigned automatically elsewhere, so that they don’t steal ports. That means administering a back-end VM would require you to use an explicit port number, but for normal users, they would access a front-end VM and just use a display number. I don’t know if that makes sense or not. This could be a separate patch too. It’s not really the same problem after all.
This default range is 5901-5999, so this would not conflict with anything running on non-system ports, unless you specify that in qemu.conf (but then you know what you are doing)
I’m not sure I understood your comment. Do you mean 4000 VMs running concurrently, or 4K display? For the latter, I don’t see how port assignment matters, so I deduce you talk about running more than 99 VMs simultaneously. Can you do that without tweaks to qemu.conf? It’s really a trade-off. Experienced sysadmins will have to extend available ports in qemu.conf (and I place anybody running more than 100 VMs on a single system in that category). Regular users won’t have conflicts with vino-server or (if über-unlucky) X11. That seems like a reasonable trade-off to me. BTW, if you really run 4K VMs on a single host, assigning one port per VM for display will become a problem anyway. There are only 64K ports available, so using 4K just for displays does not scale well, not even counting the ports you might need for the VM workloads themselves. It might be time to think about adding a VM ID directly in the protocol and having multiple connections to the same port. |
-- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list