Ever heard of the VM I/O blender? Maybe the term is new to you, but if you’ve consolidated a lot of servers using virtualization, you’ve probably felt it.
A physical server running 8 VMs can have 80% less aggregate disk I/O performance as the same box running a single OS. As you add VMs, performance degrades because the intermixed I/O loads from multiple operating systems don’t stream smoothly anymore. They swirl and churn and mix unpredictably... in the VM I/O blender.
Another common problem is snapshot performance. You build your VMs from golden images. Most of the VMs in your operation were built from a handful of golden images. Most of the bits in those images are the same, and won’t ever change. It makes sense to use snapshot technology for space efficient clones. Unfortunately, most snapshot technologies weren’t designed to perform well in a virtual server use case. How unfortunate.
It’s bad enough that performance drops. Unacceptable, really. Worse, the resultant performance isn’t predictable or reliable. So, while you know performance will get worse, you don’t know how much or when... and you’re supposed to deliver consistent quality of service?
You can’t achieve the desired level of server consolidation if the penalty is unpredictably lower performance. For good reason, IT architects are loath to virtualize I/O intensive applications.
- Storage sprawl
- Storage performance problems
- Storage management complexity
- Excessive storage costs
