A lot of great info here. Thanks very much! I very much appreciate it.
Take care,
--Al
-----Original Message-----
From: Christian Schoepplein <chris@schoeppi.net>
Sent: Thursday, July 13, 2023 7:47 AM
To: Al Puzzuoli <alpuzz@gmail.com>
Cc: debian-accessibility@lists.debian.org
Subject: Re: What's the most Accessible Linux VM Server Platform?
Hi Al,
On Wed, Jul 12, 2023 at 01:09:43PM -0400, Al Puzzuoli wrote:
I am thinking I'll run a Linux virtual machine and in that machine,
I'll run several small docker containers such as Pihole, Plex, and a few other things.
No problem. This can be done with kvm / qemu and libvirt without any problem regarding accessibility. You can setup the VM via a virtual serial console connected and perform all steps you would normaly do in the installer in this virtual terminal. Another approach is to use preseeding and let the installer perform all steps automaticaly. I am working with both aproaches very ofthen and I can provide instructions how to start the installer with serial console connected or with preseeding. Also I can provide a preseed file with all instructions for the installer.
If you have installed the virtual machine you can ssh into it and install docker and docker compose to get all applications you like to use up and running.
Thats the setup we have at work and it works very well. We install virtual linux machines more or less automated, configure them with ansible, ssh into the virtual machines and setup our docker based applications by providing the necessary docker-compose-yml files.
I'll also want to be able to run a Windows environment with at least
one virtualized domain controller, so that will be its own full virtual machine.
Thats a little bit more tricky because you can't setup the Windows VM full automated. In this case I'd do the installation manualy on a local machine, e.g. your workstation or laptop and transfer the VM to a central server later, if everything in the VM is working and properly configured. the most difficult part to install a virtual Windows with kvm / qemu is to integrate the virtio drivers. Those drivers are not included in the Windows installer per default. It is possible to setup a Windows VM without those drivers, but the virtio drivers offer the best performance. You should use the virtio drivers for the virtualized harddisk, there a scsi disk is emulated, and for the network interfaces of the VM. The drivers are all stored on a ISO file which can be downloaded and added to the VM when it is started the first time for installation. After the Windows installer has started you can start Narrator, the Microsoft screen reader, and tell the Windows installer to use the virtio drivers for the harddisk and network interface. this is the most difficult part, but if this has been done the Windows installation can be finished with the Microsoft screen reader like on a normal computer.
When the VM is up and running localy you can configure your services and also install another screen reader, e.g. NVDA or Jaws. Also you can configure the remote access via spice. Spice is the technique to connect to virtual machines running with kvm / qemu. If this is working and you can connect from your local linux machine to the still localy running VM via spice and use the screen reader inside the VM, you can stop the VM and transfer it to your server. After the transfer is done you can connect to the remote VM either by its new ip or name, or you can configure the local management software for virtual machines running with libvirt to connect also to the libvirt running daemon on your server.
I've done all those things without sigted help. The biggest issue was to get the Windows VM up and running and to include the virtio drivers, but this was more a problem because I am not so familiar with Windows, the Narrator screen reader and so on and not because it is a problem regarding accessibility. I've done all this things with Windows 10, but I think it should also work with Windows 11.
And of course you do not to use spice to connect to the remot VM, you could also use Nvda remote or the Jaws tandem technique or whatever you like and know and what is working with a screen reading software.
Do I understand correctly that there's not much of a performance hit if
you run docker containers within a VM as opposed to on a bare metal
host system?
IMHO this depends on your hardware. If your host where the virtual machines are running on is powerfull enough you will not have any trouble regarding performance. We sometimes have 20 docker deployments run inside a VM and this is OK as long as the host has enough RAM, fast disks (the best is to use ssds) and enough cpu power and as long you have configured enough ressources for the virtual machine. Especialy if you really want to use ZFS as the host filesystem the more RAM you have for the host system the better for the performance...
bTW.: We use ZFS for the data storage for our VMs and for the VMs itself we are using ext4 without LVM. If you make sure that the root file system is the last partition of all your VMs you can easily resize the harddisk, the root partition and the filesystem of the virtual machines without LVM. We do not have a seperate partition for the data stored inside VMs, just the boot partition, the efi partition and the big root partition, which is always the last partition. If you want to split your VMs into more partitions just make sure that the partition, which maybe has to be resized, is the last partition of your virtual harddisk. Also we do not use a seperate swap partition for the VMs. Either we have no swap space configured at all, or we are using a swapfile, but if a VM is swapping, just give it more RAM and make sure that no swapping is done, because swapping for a virtual system can really be a performance killer.
Ciao,
Schoepp