Determining the appropriate hardware for the compute nodes greatly depends on the workloads, number of virtual machines, and types of applications that you plan to run on the nodes. Typically, the engagement of a Mirantis cloud architect is required to properly plan the capacity for your cloud.
That said, it is essential to understand your cloud capacity utilization tendencies and patterns to plan for expansion accordingly. On one hand, planning expansion too aggressively may result in underutilization. Underestimating expansion trends, on the other hand, threatens oversubscription and eventual performance degradation.
Mirantis provides a spreadsheet with the compute node calculation. You need to fill the following parameters in the spreadsheet:
| Parameter | Description |
|---|---|
| Overhead components | Describe components that put additional overhead on system resources, such as DVR/vRouter and Hypervisor. The parameters specified in the spreadsheet represent standard workloads. The DVR / vRouter parameters represent a Compute node with 2 x 10 Gbps NICs. If you use a larger capacity network interfaces, such as 40 Gbps, this number may increase. For most deployments, the hypervisor overhead parameters equal represented numbers. |
| HW Components | Compute profile represents the hardware specification that you require for the specified number of virtual machines and the selected flavor. The adjusted version of the compute profile represents the hardware specification after correction to overhead components. |
| Oversubscription ratio | Defines the amount of virtual resources to allocate for a single physical resource entity. Oversubscription ratio highly depends on the workloads that you plan to run in your cloud. For example, Mirantis recommends to allocate 8 vCPU per 1 hyper-thread CPU, as well as 1:1 ratio for both memory and disk for standard workloads, such as web application development environments. If you plan to run higher CPU utilization workloads, you may need to decrease CPU ratio down to 1:1. |
| Flavor definitions | Defines a virtual machine flavor that you plan to use in your deployment. The flavor depends on the workloads that you plan to run. In the spreadsheet, the OpenStack medium virtual machine is provided as an example. |
| Flavor totals | Defines the final hardware requirements based on specified parameters. Depending on the number and the virtual machine flavor, you get the number of compute nodes (numHosts) with the hardware characteristics. |
| Resource utilization per compute node | The resource utilization parameter defines the percentage of memory, processing, and storage resource utilization on each compute node. Mirantis recommends that vCPU, vMEM, and vDISK are utilized at least at 50 %, so that your compute nodes are properly balanced. If your calculation results in less than 50 % utilization, adjust the numbers to use the resources more efficiently. |
See also
Download Compute nodes calculation