It hasn’t been said directly but the clue was in the Amazon’s post “Reserved Instance prices dropping by as much as 37%, and On-Demand instance prices dropping up to 10%”. The relation between on-demand and reserved instance prices was not kept and it means the switch point between them has moved as well.
The point when it is more cost effective to use reserved instances differs between instance types but in general it was about 30% utilization during one year reservation period, which is equivalent to 2630 hours. For standard and high CPU instances it was 31.8% (2786 hours), while 29.2% (2559 hours) for high memory instances. These are the points where you should switch from on-demand to one year light reserved instances in US East
With the new price lists the most affected were standard and high memory instances. The profitability point of the standard reserved instances moved from 31.8% to 19.2% (1683 hours). For high memory instances the change was from 29.2% to 17.5% (1535 hours). High CPU instances had just minor update from 31.8% to 31.3%. Cluster compute and GPU instances did not change the relation, but keep in mind Cluster Compute Eight Extra Large is still the winner with just 13.4% utilization (1178 hours) for the light reserved instance to be profitable. The profitability points for other types of reservations also changed but not that much. The maximum is from 76.9% to 65.2% for high memory medium utilization reserved instance. The heavy utilization changed by 2% at most.
But what does it mean to me? If you were below the threshold so far and you use on-demand instances, you should definitely check if now it isn’t cheaper to take light reserved ones. If your server runs for more than 4 hours 12 minutes a day – check reserved instances. If you know your server will run at least 2 months and 1 week all the time, you are again in the reserved instances zone. Similarly for 9 months of 8 hours per business day, or any other pattern in which you would exceed the required number of computing hours. Similarly you should also check if a higher reservation wouldn’t provide a better deal. If you use the cloud computing comparison engine, it will tell if and which reservation is most appropriate in a given case.
To get maximum of the cloud economics you should always check if you use the best option. Currently many users will find profitable to either switch from on-demand to light reserved instances or going to higher reservation level. Check if you are one of them.
Profitability points for 1 year reservations in US East:
Light Reservation | Medium Reservation | Heavy Reservation | ||||
Old | New | Old | New | Old | New | |
| Standard | 31.8% 2,786h | 19.2% 1,683h | 74.2% 6,500h | 69,3% 6,067h | 85.2% 7,466h | 83.3% 7,299h |
| High Memory | 29.2% 2,559h | 17.5% 1,535h | 76.9% 6,740h | 65.2% 5,713h | 85.2% 7,467h | 81.6% 7,150h |
| High CPU | 31.8% 2,786h | 31.3% 2,739h | 74.2% 6,500h | 67.6% 5,925h | 85.3% 7,474h | 82.8% 7,257h |
| Cluster Compute Quadruple XL | 29.7% 2,599h | 29.7% 2,599h | 71.8% 6,288h | 71.8% 6,288h | 85.6% 7,502h | 85.6% 7,502h |
| Cluster Compute Eight XL | 13.4% 1,178h | 13.4% 1,178h | 74.8% 6,550h | 74.8% 6,550h | 84.9% 7,438h | 84.9% 7,438h |
| Cluster GPU Quadruple XL | 31.8% 2,783h | 31.8% 2,783h | 74.4% 6,519h | 74.4% 6,519h | 85.3% 7,470h | 85.3% 7,470h |
