Database Reference
In-Depth Information
not be stepping in all the time. SIOC should be seen as more of a last resort. If you
observe higher latency in your VMs and SIOC working constantly, this is an indication
that your data store or storage platform can't support the required IO workload. You
may need to add more physical disks to your storage platform or reorganize some of the
LUNs to reduce hot spots.
As shown in
Figure 6.28
,
if one VM or one host begins to monopolize the available
performance resources, the other VMs sharing the same data store or storage array
suffer.
Figure 6.28
The Noisy Neighbor Effect.
In some cases, it's not just that other VM's performance suffers, but other more
important VMs sharing the same data store don't get the IO resources they are entitled
to.
Figure 6.29
provides an example where three VMs share the same data store. One
important VM and a less important VM share a vSphere host, while another less
important VM is on another vSphere host. The relative importance is defined by the
shares value, which uses a proportional share algorithm to carve up the performance
resources. Because this doesn't work across hosts, the less important VM on its own
host has full access to the available queue depth and therefore is getting more than its
fair share of IO performance resources.
