Challenges in benchmarking
virtual environments
Clock skew – for benchmarking in
a virtual environment the clock can actually be impacted or not the same and
skew some of the metrics. If metrics are collected by the hypervisor there is
less chance of this happening.
NUMA - (non-uniform memory access) scheduling. CPU’s which may be
assigned to certain sections of memory and when you allocate numerous machines
and they go across different spaces you can end up with some performance
degradation.
Pass-through I/O - provides disk
performance on a par with physical disk performance. Pass-through disks are
essentially physical disks/LUNs that are attached to a virtual machine and do
not support some of the functionality of virtual disks. All the I/O that I am
doing is virtualized.
Workload definition – You would normally
wish to test or benchmark with your own workloads.
32 versus 64 bit OS and applications – understanding hardware and software system requirements
Where to use solid state I/O devices – solid state I/O devices are becoming more popular so you will need to
know when and where to use these.
Storage tiers – based on usage and priority policies you need to look at setting
a strategy for for tiered
storage.
Hidden overhead – Virtualization
requires overhead so you do need to keep this in mind.
The
list could go on and on…………but I have covered what I consider to be some of the main challenges.
On Friday I'll be telling you the test environment and testing methods that I used to compare performance.
Dale Feiste
Consultant
No comments:
Post a Comment