Overallocated VM recommendation
The Overallocated VM recommendation identifies virtual machines (VMs) with underutilized resources (CPU, memory, or storage). A VM is detected to be overallocated if the spare capacity of its resources (CPU, memory, storage) exceeds the corresponding threshold. The spare capacity of a resource is the difference between its actual configured value and its estimated future usage value.
The Overallocated VM recommendation is available for these technologies: AWS, Azure, Hyper-V, OpenStack, and vSphere
The following recommendation details are provided only when you apply Service Pack 1 (11.5.01) of TrueSight Capacity Optimization 11.5:
- Recommended action that suggests reconfiguring the VM with the ideal configuration values to save costs. For a source VM that is in the public cloud, the recommended instance type for the target VM belongs to the same family as that of the source. For example, if the source VM is of type General Purpose - M4, then the recommended target instance can belong to M3 or M5, depending on the computed ideal size.
- Comparison showing the difference between the current utilization of VM resources and their utilization after reconfiguring the VM
- Criteria for detecting the VM as overallocated
- Benefits of implementing the recommendation
You can analyse these recommendation details and take any of these actions:
- Reconfigure the VM with the recommended values to reclaim storage and achieve cost savings
- Increase the VM load by allocating additional applications to improve the VM capacity utilization
The recommendation rule is based on these thresholds in the Optimizer rule:
- Spare CPU (Cores)
- Spare Memory (GB)
- Spare Storage (GB)
You can change the default values of these thresholds on the Rules page in the TrueSight Capacity Optimization console.
For more information about modifying the Optimizer rules, see Adding a template-based Optimizer rule.
The Memory Demand indicator is used to determine whether a VM is overallocated. For vSphere VMs, this indicator is computed by considering the 95th percentile value of the MEM_REAL_UTIL or MEM_ACTIVE metrics. When using the MEM_ACTIVE metric, the vSphere VM Active Memory increase factor is applied to increase the MEM_ACTIVE metric value. For more information about the Memory Demand indicator, see demand indicator.
You can change the value of the vSphere VM Active Memory increase factor on the Thresholds page in the TrueSight Capacity Optimization console.
For more information about configuring thresholds, see Configuring and managing thresholds for metrics and indicators.
The recommendation is assigned an efficiency level (high, medium, or low) based on the amount of estimated cost savings per month.
Efficiency level | Estimated savings amount (in $) per month |
|---|---|
high | $50 or more |
medium | Between $25 - $49 |
low | Less than $25 |
You can modify the efficiency levels on the Rules page in the TrueSight Capacity Optimization console.
For more information about modifying the Optimizer rules, see Adding a template-based Optimizer rule.
Ideal size computation of the target VM
The ideal size of the target VM is calculated by considering the estimated future usage and the estimated past usage of the resources. The estimated past usage value of each resource is computed based on the selected Optimization behavior (Aggressive, Balanced, Conservative) and stored in the corresponding demand indicator.
For VMs that are provisioned in a public cloud, the resizing recommendation for the CPU resource also considers CPU benchmark. The estimated CPU benchmark value of the source VM is compared with that of the probable target VMs and the VM with the best match is recommended.
If the CPU benchmark value for the source VM is unavailable, then the resizing recommendation is generated without considering benchmarks. SPECint_rate2006 is used as the reference benchmark to compare the hardware of the source and target VMs. The CPU benchmark consideration is to ensure that the suggested target VM performs at the same computing power as that of the source VM, even if the hardware is different.
Past usage estimation
The following table explains how the past usage is computed per optimization behavior:
Optimization behavior | Description |
|---|---|
Aggressive | Resource utilization in the server is computed by considering the average value of the hourly samples. Then, 95th percentile of the hourly value over the last 30 days is computed for each resource to generate the configuration of the target virtual machine or instance type. Spikes in the resource utilization within the hour are not considered. This optimization behavior does not require the server to be instrumented because granular or detailed metrics are not used for the computation. |
Balanced (Default) | CPU utilization in the server is computed by considering the 90th percentile value of hourly samples. Then, 95th percentile of the hourly value over the last 30 days is computed for CPU to generate the configuration of the target virtual machine or instance type. 90% of spikes in the CPU utilization within the hour are considered. Utilization of memory and storage is computed by considering the 95th percentile of the average hourly values over the last 30 days. This optimization behavior requires the server to be instrumented to collect the granular or detailed metrics. For VMware vSphere, the server need not be instrumented because the vSphere Service ETL collects data at less than one minute granularity. |
Conservative | CPU utilization in the server is computed by considering the 95th percentile value of hourly samples. Then, 95th percentile of the hourly value over the last 30 days is computed for CPU to generate the configuration of the target virtual machine or instance type. 95% of spikes in the CPU utilization within the hour are considered. Utilization of memory and storage is computed by considering the 95th percentile of the average hourly values over the last 30 days. This optimization behavior requires the server to be instrumented to collect the granular or detailed metrics. For VMware vSphere, the server need not be instrumented because the vSphere Service ETL collects data at less than one minute granularity. |
The computed resource utilization values are stored in the demand indicators (CPU Demand, Memory Demand, and Storage Demand). For more information, see Indicators.
For information about the data sources that enable you to collect metrics for the optimization behaviors, see Data-source-requirements-for-VM-resizing.
Other recommendation details
The Overallocated VM recommendation window displays an Accuracy icon that indicates the accuracy or reliability of the resizing suggestion given in the recommended actions. The accuracy level shown in the icon is based on the selected Optimization behavior. For more information about the accuracy levels, see Accuracy-levels-of-resizing-recommendations-for-overallocated-VMs.
Where to go from here
For instructions about viewing the overallocated VM recommendations, see Analyzing-the-overallocated-VMs-in-your-multi-cloud-infrastructure.
Related topics
Optimizing-multi-cloud-costs-with-out-of-the-box-recommendations