You are here

Home » Benchmarks

Amazon EC2 Benchmarks


After I have started to compose SSD VPS performance chart, I was contacted several times about building similar chart for Amazon EC2 instances. Below is the first version of that chart.

I added also the projected monthly cost for every instance. Note that this only includes the cost of EC2 instance itself, bandwidth and EBS volumes charges are added, wherever applicable. Please refer to the Amazon EC2 reference pages for more details.

In case of 1-year and 3-year reserved instances you pay some sum upfront, after that the hourly cost of the affected EC2 instance becomes less. The monthly cost in the chart is an average cost, all the hourly and upfront expenses taken together and distributed over the span of the reserved period. So do not get confused when you will see in your Amazon monthly statement different sums, the below prices are given to give an impression of the lowest sum you would pay.

Prices are for US-East region. Prices for other regions might be higher. I can do all the charts for all the regions if there's significant amount of inquires to do that.

Amazon EC2 performance chart

The methods of calculations and software used is the same used to measure SSD VPS performance. You can download the archive containing all the benchmark data obtained, for more detailed study.

Note that IOPS parameters gathered aren't very consistent. I am now testing them in many passes, doing long pauses between subsequent calls and altering read and write tests calls, to both have the system settle after the previous run and get more stats to average. The gathered IOPS data were not calculated in that way and thus are subject to re-calculating in the foreseeable future. I can only add that tests of my long-running Micro instance demonstrate better IOPS than those in the table.

Code ECU Cores RAM Price, USD/month UB IOPS
Linux Windows
hourly 1 year 3 years hourly 1 year 3 years Read Write
t1.micro1-21613 Mb14.6112.8711.9321.9215.0714.12115.09061463
m1.small111.7 Gb58.4434.2425.6084.0148.8540.21185.2133452613
m1.medium213.7 Gb116.8868.4851.92168.0297.7081.14384.3130052812
m1.large427.5 Gb233.76136.96102.39336.03194.67160.83705.14831492
m1.xlarge8415 Gb467.52273.92204.79672.06389.34321.671089.245562791
m2.xlarge6.5217.1 Gb328.73190.13143.06416.39241.26194.19889.5105582311
m2.2xlarge13434.2 Gb657.45380.25286.12832.77482.52388.391385.2223212287
m2.4xlarge26868.4 Gb1,314.90760.51572.241,665.54965.05776.782273.7138342107
c1.medium521.7 Gb120.5387.8871.87208.19135.37119.35651.9213972356
c1.xlarge2087 Gb482.13351.53287.47832.77541.46477.401657.7152702165

Note: yes, I also see strangely low IOPS random read results of Large and Extra Large types of instances and will perhaps re-calculate those at first opportunity. Looks like the nearby 5-digit values are more realistic.