Database Reference
In-Depth Information
Figure 6.24
shows the different areas where storage IO latency is measured and the
relevant values inside vSphere. DAVG, which is the device latency, will indicate if you
have a bottleneck in your storage array, which may mean you need to add more disks or
reduce the load on that device. If you start to see KAVG constantly above 0.1ms, this
means the vSphere kernel is queuing IOs and you may need to increase device queue
depth, especially if the DAVG is still reasonable (< 10ms).
Figure 6.24
VMware vSphere storage latency.
We want to optimize the queues through the IO stack so that the disk devices are the
constraint, and not the software or queues higher in the stack. Periodic spikes in DAVG
and KAVG are acceptable, provided the averages are not consistently high. Brief spikes
in DAVG and KAVG are acceptable; however, high average values are a sign of a
performance problem. Suggested thresholds are listed in
Chapter 10
.
Tip
When consolidating multiple SQL servers onto fewer hosts, there is usually an
implicit assumption that SQL was not previously making full or optimal use of all
of the system resources. This includes CPU and RAM, but also storage IO and
HBA queues. It's your job as the architect or admin of the environment to ensure
your destination vSphere platform and each host has in aggregate sufficient
resources to service the blended peak IO workloads of all of the databases on a
