At the moment I am searching for suitable servers for the infrastructure of my project. I settled on EC2 and decided to analyze which solution would be the most cost-effective for me. This article is essentially a continuation of
AWS Insight: Reservations in EC2 - how it works , but gives an analysis for a young web startup (24x7).
Problem
The server should work 24 hours a day for X months (the number is not defined). Which option (On Demand / Light Utilization Reserved / Medium Utilization Reserved / Heavy Utilization Reserved) to choose for Micro / Small / Medium / Large servers?
')
Foreword
What is the difference between the choice? Only in price. You get exactly the same server, but at different prices depending on the period during which you plan to use it.
On Demand - payment is fully hourly (high), no prepayments.
Heavy Utilization, Medium Utilization, Light Utilization Reserved - Amazon recommends using these types if your server is running continuously (100%), often enough (60%) or occasionally (40%), respectively. To reserve a server, you make a non-refundable prepayment of 1 or 3 years. After that, the charge per hour is significantly reduced. It is important to note that in the case of Medium and Light, only those hours in which the server was used are paid. Heavy - all hours are paid until the end of the reservation, regardless of server usage.
Analysis
For brevity, I will omit the whole analysis, but I will show its principle. For each type (Micro / Small / Medium / Large), I compiled a table of the type shown below for Large. The cell shows the full amount Amazon will get from you after a certain month. Green marked the smallest amount. Red - the greatest.

You can simplify this picture with the following table. The cell contains the plan with the smallest amount you pay after the month has passed.

It's funny that above we do not see a single Medium. Is Medium such a loser for this Use Case? Not. If we think for a second, we will understand that the above tables do not answer unambiguously the question posed. The fact is that it has already been noted earlier: Heavy Utilization obliges you to pay all, even unused, hours for the entire reservation period.
Thus, we are not so much interested in how much we will give to Amazon after a certain month, but in how much we will be obliged to give everything, taking into account that the server will not be used after a certain month.
Change the auxiliary table so that it shows the real cost of the server after a specific month.

The last two columns changed, but made quite a lot of red and green cells change their positions. We summarize:

Ta-dam! And here is the Medium!
findings
So now the fun part. What did we get?
Since I am interested in Large Instance, I will make a start from it. Although the trend for all types of servers is the same (offset + - 1 month).
- Perhaps the most interesting is that Heavy Utilization 1 year did not justify its descriptions. He beats competitors only in the last two months (11 and 12) of his contract.
- But Light Utilization 3 years seems very attractive, because it covers as much as 8 months (4 to Heavy, 4 after), which gives not only the opportunity to extend the economic use up to 16 months, but also the ability to use it for its intended purpose (20-40%) for the next 20 months.
- Light Utilization 1 year is most beneficial for projects from 3 to 6 months
- It was not surprising that On Demand solution is the most cost-effective in the first 3 months of operation, which is 2200 hours. Monthly price is fixed and is equal to $ 190.32
- Medium Utilization 3 years and Heavy Utilization 3 years are struggling for the second half of the table. Medium significantly reduces the risk of expenses, as it covers 33% (17-28 months) of the table. And, of course, the flexibility of payment per hour makes it a worthy candidate for choice.
- Heavy Utilization 3 years looks much more worthy than its one-year-old fellow. If you need a server for 28 months or more, then without question. The amount of the question - $ 2,570.30
- Medium Utilization 1 year - the greatest disappointment for this issue. Won one small nomination for Micro. Unfortunately, not worthy of consideration. Sorry friend!
Links
The remaining auxiliary tables are
on Google.