HP Capacity Advisor 7.2 User Guide

Data seems to be incorrect or lost
VM guest resource utilization appears to exceed the resource utilization of its
VM host
When viewing historical resource utilization data in a profile viewer, it can appear that the utilization
of a resource by a given VM guest exceeds the utilization of that resource by its VM host at a given
instance of time. (For example, you might see the I/O usage occasionally exceed the VM host's
high-water mark value (the dashed blue line in a profile viewer graph).
In reality, it should never be possible for a VM guest's resource utilization to exceed that of its host,
since the utilization shown for a VM host is the aggregation of a resource's utilization for all VM
guests running on that host. Yet this does occur on rare occasions due to differences in how the
data was collected, interpolated, and averaged into the five–minute intervals that Capacity Advisor
stores in its database.
Such anomalies become insignificant when the data is averaged over sample intervals greater
than five minutes.
Note that this situation typically occurs only when a single guest is running on the host and is
creating a very large load on the host.
Peak utilization in a scenario is greater than 100%
Though this situation is rare when you first create a scenario, it can occur in the following situations:
• When virtual machines are modeled to be using a lot of I/O bandwidth, because some
hypervisors will give the VM more CPU time than seems possible given the virtual processors
assigned to the VM.
• When forecasting growth of workloads into the future. Over time, the forecast model will show
workloads growing beyond the size of their hosts.
• When a real system simulated in a scenario is down-sized during the useful life of a scenario.
The scenario will retain the original description of the system, but will be plotting data captured
from the smaller real system.
As you begin to edit the systems in the scenario, you can easily edit attributes in a way that will
cause the simulated data to exceed 100% utilization. Use the data displays in the scenario editor
to help you check the validity of the simulation you have constructed, and watch out for the following
situations:
• When editing simulation system attributes, or when importing data using capprofile, be
aware of discrepancies in clock speed or memory between the real system and the simulated
system. To associate the characteristics of the real system from which data is being imported
to the simulation system, use the -o -S options with the capprofile command.
• When converting standalone servers to VMs, be sure to include an appropriate virtualization
overhead in the utilization calculation for the new VMs. (See CPU Virtualization Overhead %
for more information on this calculation.)
Socket count is missing or seems incomplete in population report
Socket information is not always available due to limitations in data collection from some platforms
or when leveraging data collection techniques that do not include socket count in their collection.
There is no remedy in these situations.
Workload performance seems inaccurate
Using benchmarks for data normalization affects workloads defined with static profiles in older
Capacity Advisor releases (prior to 6.0) because they use clock speed to determine performance.
Suggested action: Such workloads that are still used for planning should be redefined or edited in
Data seems to be incorrect or lost 163