Auto-scale: Reserved Instances Analytics
Reserved Instances (RI) are the ability to prepay, for a term or 1 or 3 years, for a certain amount of CPU cores in a certain Azure region, in a certain VM family. This gives you a significant discount on the cost of compute. Typically, the average discount for a 1-year reservation is about 40% and a 3-year reservation is about 60%, compared to pay-as-you-go pricing.
It rarely makes sense to purchase RI for all of your hosts in all of the host pools, rather than using auto-scaling. Even with a 3-year reservation savings of 60%, that's still less than the typical 70%(or more) savings from auto-scaling. This is because the amount of time the VM is actually used during a typical week is generally below 30%. For example, if the VM is on for 40 hours during the week, out of 168 total hours in a week, it is only in use only 24% of the time. That means auto-scaling would give approximately a 75% savings.
That said, even if you are using auto-scaling, it makes sense to purchase RI for a certain number of CPU cores. Calculating how many cores to purchase is a complicated analytics exercise. Nerdio Manager provides a simple guide that looks at the auto-scale history of a particular host pool and provides you with RI recommendations.
Tip: Auto-scaling typically reduces the number of cores needed, for example, from 100 cores to 25 cores. (Based on a typical 40-hour week.) Those 25 cores can then be reserved to reduce the cost from 40%-60%, based on the reservation's term. Bottom line, Nerdio Manager reduces the cost of compute by 80%-85% compared to pay-as-you-go pricing.
To generate RI recommendations:
Locate the dynamic host pool you wish to analyze.
From the action menu, select Auto-scale > RI analytics.
Select the date range to use for your analysis. The metrics and recommendations are based on this date range.
The following metrics and recommendations are shown:
CPU-hours for select dates: For the selected date range, this is the total number of hours that the VMs were powered on multiplied by the number of CPU cores in each VM.
Projected CPU-months: This is a projection of how many CPU cores will be fully consumed on a monthly basis based on the behavior in the selected date range. (This is based on 730 hours in a month.)
Based on the metrics, Nerdio Manager provides RI recommendations for the Region, VM family, and Total CPU cores.
Tip: You should purchase this number of Reserved Instances CPU cores, in any combination of VM size, within the recommended VM family and Azure region.
CPU cores example:
If Nerdio Manager recommends that you purchase 100 cores, you can purchase these cores in any VM size as long as it is within the recommended Region and VM family. So, you can purchase 3 32-core VMs and 1 4-core VM, for a total of 100 cores.