Azure offers instance reservation for 1 or for 3 years with a great discount that you shouldn’t miss.
PAY as You Go prices is good for short term, spiky usage, while reservation is fit for long term 7x24 hours running virtual machines.
What is the challenge that who has solid view what will happen in 3 years, meaning application architecture unchanged, capacity will be only increasing, etc.
It is better as it seems, about 1.5-2 years forecast is enough, let us see the details.
Firstly, let me highlight that only the 7x24 running instances are worth to reserve. If VM can be shut down during night or for the weekends, it can be 70% cost saving without reservation, never forget.
Back to reservation, let us calculate the total cost of ownership comparing options for
Below chart shows the TCO for each month in case of a D16s v3 CentOS/Ubuntu instance located in East US.
The dashed red line shows the lowest TCO.
Typically, the VM lifespan can be estimated only with likelihood. There is a possible pitfall here, that VM is reserved for 1 year and after 12 months it is reserved for additional 1 year which is the same cost as it would have been reserved for 3 years originally.
Similarly, there is a risk that after 19 months the instance is not decommissioned as a project may be delayed or whatever reason. Therefore, it could be considered to reserve the instance for 3 years if the instance is expected to be running for more than 14 months where 3 years reservation TCO is smaller than PAYGO.
The reservation rule depends on how long the instance possible to be running:
The 7,14 and 19 “magic numbers” varies on the instance type prices, practically depends on the RI discounts, easily calculated with the formulas below.
Risk can be handled by having a “portfolio”, keeping x% of the instances with 3 years reservation while let other fraction reserving for 1 year and the remaining for PAYGO.
PAY as You Go prices is good for short term, spiky usage, while reservation is fit for long term 7x24 hours running virtual machines.
What is the challenge that who has solid view what will happen in 3 years, meaning application architecture unchanged, capacity will be only increasing, etc.
It is better as it seems, about 1.5-2 years forecast is enough, let us see the details.
Firstly, let me highlight that only the 7x24 running instances are worth to reserve. If VM can be shut down during night or for the weekends, it can be 70% cost saving without reservation, never forget.
Back to reservation, let us calculate the total cost of ownership comparing options for
- Pay as you go prices
- 1 year reservation iteratively
- 1 year reservation and pay as you go
- 3 years reservation
Below chart shows the TCO for each month in case of a D16s v3 CentOS/Ubuntu instance located in East US.
The dashed red line shows the lowest TCO.
Typically, the VM lifespan can be estimated only with likelihood. There is a possible pitfall here, that VM is reserved for 1 year and after 12 months it is reserved for additional 1 year which is the same cost as it would have been reserved for 3 years originally.
Similarly, there is a risk that after 19 months the instance is not decommissioned as a project may be delayed or whatever reason. Therefore, it could be considered to reserve the instance for 3 years if the instance is expected to be running for more than 14 months where 3 years reservation TCO is smaller than PAYGO.
The reservation rule depends on how long the instance possible to be running:
- Less than 7 months, keep it with PAYGO
- Running between 7 - 14 months, reserve for 1 year and keep PAYGO
- Running more than 14 months, reserve for 3 years.
The 7,14 and 19 “magic numbers” varies on the instance type prices, practically depends on the RI discounts, easily calculated with the formulas below.
Risk can be handled by having a “portfolio”, keeping x% of the instances with 3 years reservation while let other fraction reserving for 1 year and the remaining for PAYGO.
- SAP Managed Tags:
