The great idea behind cloud computing is the separation of the software and hardware. Before cloud, in order to put up a website, one would buy a server, have it setup and OS installed by a technician, and send it off to the data-center to be tucked into a rack with power, network, and cooling. Now software designers can concentrate on the OS and their software while GMO Cloud provides and takes care of the physical hardware.
So GMO Cloud makes sure that the physical hardware underlying the virtual machines keeps running. If that is working well, accessibility and cloud flexibility services to the virtual machines running on them can be provided - services not available to those running their own hardware and software.
The virtual machines the customer uses acts just like a physical machine. This is so the customer can install the server OS they need and everything performs as it should. So the customer manages everything inside that virtual machine. But wrong configurations, user mistakes, malicious code, bad updates, traffic spikes, run-away processes, etc do still happen inside the virtual machine. Therefore the user knows best what to monitor and how to fix it.
Monitoring alerts and such give the user more control and warning without having to constantly watch the server themselves. For most situations the user can decide what should be done and they are given the tools available to a cloud service.
When the situation arises where the problem is in the provider's physical hardware or cloud services and not the user's virtual machine, there are methods for solving those:
Physical Hypervisor failure: This is automatically detected on the provider side and the virtual machines are moved to another hypervisor.
Damage to the virtual machine directory: Some major change to the virtual machine disk directory, kernel, or such. This is where automated backups help.
Performance de-gradation: Hot/Cold migration allows the user to test out on another hypervisor.
Minimizing effects of hypervisor crash: Separating the virtual machines by migrating to separate hypervisors makes this possible.