The other case for virtualization

Cloning Experiments: Jess Payne
Originally uploaded by orangeacid.

We all know that by virtualizing our servers, we reduce hardware and power costs. When looking up information about server virtualization, that's 99% of the reason to do it. There are other and even better reasons, from the stand point of a system administrator, to virtualize.

Where I work, we have many servers that we put into virtual environments. We have a pretty good balance between virtual and iron. For some tasks, people are just not comfortable using a virtual environment like Virtual Center for things like Nagios (monitoring) or Cacti (metrics), so these servers get the bare metal. The IO some servers consume make sharing hardware just not worth it.

Now even though some servers deserve their own hardware, they can still be virtual servers on that hardware. Even if it's a single virtual server on a single hardware platform (though running two would be nice so you can have a fail-over cluster).

Portability - If all servers are virtual, even the single ones (one virtual to one hardware platform), then upgrading hardware is painless. The server and hardware are no longer attached so just about anything with room and power can house this server.

Server builds - all of the hardware can now be installed using pretty much the same template. All you do is rack up hardware and build it as a dom0 (or host). Then you can just copy or build the domU (guest) to the new hardware.

Remote Console - With virtual servers, you can now just ssh to dom0 and then bring the virtual servers to single user mode, to do maintenance.

Lights Out Management - Along with the remote console, you also gain the ability to remotely power off the server or reboot in case of kernel panic. This is fantastic if you have remote sites.

Redundancy (with clustering) - Redundant power supplies, dual NICs, RAID... We put a lot of money into making servers safe from hardware failure, and maybe because of this I don't really ever have a server die on me due to hardware. Oh, I have to replace hard drives frequently, but I never had that drop a server on me. What I -do- run into are kernel panics killing the server. In fact, most of my server downtime is caused by software issues.

Cheap STONITH - If you cluster two together on the same system, you can "Shoot The Other Node In The Head" via dom0 so you save on having to buy special power strips.

I know that this makes the server more complex, and you can also run into problems on the base OS, but if you really tailor that build to be dedicated to just running virtual servers, you really minimize that risk. It's really the same risk as just running the one OS for the one hardware platform.