Which way to achieve fault tolerance to choose?

Question

High availability

Good day to all. I have a problem that, due to inexperience, I don’t understand how to solve, so I demand a strategic kick in the right direction.
There is a system that communicates with the outside world via USB(Com-port), RS-232 and Ethernet. At the moment, everything is spinning on several PCs with Windows 7, on which non-clustered software is installed, which requires the physical presence of USB-HASP (does not understand over the network). The software works with MS SQL Server 2008 R2. Each PC has Win7, software, MS SQL, Hasp key and several physical interfaces.
At the moment the system works, but it is very unstable. So unstable that each user has a pack of instructions for eliminating all sorts of glitches, which he successfully uses ... every day.
The main part of failures falls on USB, RS-232 devices/interfaces. The mechanism for replacing these nodes with similar, much more stable ones, but with Ethernet, has already been worked out. True, you have to involve Debian, Dos, Asterisk.
At the same time, it was decided to spin all this on a failover cluster of virtual machines on two existing Dell PowerEdge R810, upgraded to the required configuration with minimal investment. Paid software for clustering is considered very reluctantly due to the fact that it is not clear when it will beat itself off.
At the moment, Proxmox 5 + GlusterFS is deployed on two existing nodes (in testing mode) as a separate link. To ensure a quorum, it is planned to use the third node on a low-performance PC, OrangePi or something similar that will pass combat tests. So far, this node is using a VirtualBox virtual machine on a working PC with 8 GB HDD, 512 MB RAM, debian 9 + proxmox (without connecting to GlusterFS).
While everything is working fine, Windows 7 virtual machines successfully migrate online in 5-10 minutes, successfully migrate when one of the nodes crashes, BUT ... with a reboot nodes and, as I was able to find out, even what it can do for money does not do it confidently enough.
What if this five-minute downtime is highly undesirable? The main problem with Windows, a five-minute downtime on secondary OS is acceptable. I suppose that you need to organize two virtual machines with Windows Server 2008 R2 (into a cluster of them?), which will be raised on two nodes. On both virtual machines, raise MS SQL in a cluster configuration and somehow organize the migration of the IP-shniks of these virtual machines from one to another, while one of them must sleep when the other is working (active / passive), otherwise the database will be a mess due to lack of communication with equipment on one of the machines. The software acts only as an intermediary between the base and interfaces, and the entire state of the software is determined by the data in the database, so connecting to the base and interfaces of another instance of the software will not lead to anything bad.
What technologies to read? What to use? It is unlikely that anyone will give money for a separate storage, so we should try to do without it.
The question with HASP keys is also of interest in this case. Should I bother with forwarding these USBs using Debian + OrangePi, or should I immediately buy something like AnywhereUSB? Maybe there are some other options?
I ask you not to kick too much, since this is essentially my first serious project, there is no one to ask, and if everything went quite briskly with Linux clustering, then something completely bad with Windows in this regard. I do not ask you to set up instead of me, just tell me how to do it.

Answer 1

I have 3 nodes on VE 5 external storage on iscsi uplink 10Gb/s, migration goes without reboot. virtual machines 2008r2 Terminal server, two AD, MSSql database server, 1c server and PostgresQL, not counting zabbix asterisks and other things. Better hardware AnyWareUSB