[kwlug-disc] Virtualization allocation?
rarsa at yahoo.com
Sat Nov 6 12:34:42 EDT 2010
--- On Sat, 11/6/10, Paul Nijjar <paul_nijjar at yahoo.ca> wrote:
> For those of you who do Linux server virtualization: how do
> you split up the load?
We use virtualization extensively at work, so even though is not specifically Linux virtualization I can provide some background. This background may or may not apply to your environment.
We started virtualizing test boxes mainly because of the following
- Faster procurement and repurposing. As long as we had the memory and CPU capacity, we could just spawn new virtuals.
- Easier test bed roll back: If we are testing an upgrade, for example, we could try it and if we found we did something wrong, we could just restore the initial snapshot.
- Easier cloning: e.g. Easier to deploy identical servers.
Of course, as John pointed out, it soon got out of control and more strict controls were required to create and dispose of test virtuals.
After that we went to production for different reasons:
- Better capacity utilization: These days it is not cost effective to procure a low power server for a purpose specific application, so we had servers running at 5% capacity.
- Flexible response to peak production requirements: We can just set-up additional virtuals for peak season, and drop them after.
- Better change management: Our change management process require coordination and approval from the affected teams. If we have multiple applications on a single server, that coordination is a nightmare. Having independent servers reduces it.
- Better disaster recovery: This is a big one. Our current disaster recovery plans require backup servers at a different location. We normally use those servers for UAT but the recovery requires more steps. Keeping an image of your production server and just restoring it at a different location is faster.
- Easier hardware currency: If we keep old servers we increase the risk of failure. We have a 5 years refresh cycle for hardware, moving applications to new hardware requires substantial planning and testing ($$$). With virtuals, we can refresh the host with minimal impact to the virtuals.
- I am sure that heat, space and power consumption are also important factors,
Cost is just a marginal benefit. The cost of the hardware is a fraction of the cost of maintaining that server, and that maintenance is the same for physical or virtuals.
Software, Hardware and Practices
An eclectic collection of random thoughts
More information about the kwlug-disc