In a previous post I compared the cost of building a private virtualized development environment in house versus deploying the same infrastructure on Amazon EC2. Given a simplistic set of assumptions about usage patterns and resource requirements, my conclusion was that building a bare-bones private cloud was about half the cost over 5 years versus building on Amazon.
One reader asked, “If cost is 2x at 30% server utilization, sounds like if the server is used 15% of the time the cost will be the same either in house or on Amazon.”
This would be correct if you ignore that at 15% utilization (25 hours per week) you would never save enough on the hourly usage rates to break even on the reservation fee. The graphic below shows the break-even cost points for EC2 reserved versus on-demand m1.large instances given various assumptions about the decrease in future average hourly usage pricing. The steeper lines are on-demand instances and the less-steep lines are reserved instances, assuming a 3-year reservation.
The line colors represent the different assumptions about average annual compound usage pricing reduction per year. Blue represents a 4% annual price decrease; yellow represents 6%; red is 8% and green is 10%.
I generated this chart in Google Spreadsheets. Grab a copy to play with if you like.
Key takeaways:
- If an instance needs to be up for less than ~48 hours per week, or around 22% utilization, you should definitely choose the on-demand option.
- If you are sure an instance will need to be up for more than ~56 hours per week, you should buy down the usage pricing by reserving your instance.
- Between 48 and 56 hours is a toss-up, and whatever assumption you make about future price reduction probably won’t make much of a difference. If you think your utilization might to grow over time, it might be best to go with reserved instances.
