Extensive Benchmarks Of Amazon's EC2 Compute Cloud Performance

Smallpt is a very lightweight path-tracing application, but like C-Ray it works very well for highlighting the differences between cloud computing platforms.

Going off the Smallpt results as a near best-case scenario for compute scaling across the EC2 instances, the m2.4xlarge instance was about 25x faster than the m1.small. This is close to the rated difference with m1.small having just one EC2 Compute Unit while m2.4xlarge has 26 EC2 Compute Units. The cost difference between these instances is also near linear at being 23.5x more expensive. For the c1.xlarge instance, it was 21x greater for 20x more EC2 Compute Units while the cost was just 8x higher. If breaking down Amazon's EC2 pricing solely in relation to the number of EC2 Compute Units (and ignoring memory or storage differences) it comes down to:

m1.small: $0.085 per hour per EC2 Compute Unit
m1.large: $0.085 per hour per EC2 Compute Unit
m1.xlarge: $0.085 per hour per EC2 Compute Unit
m2.xlarge: $0.076 per hour per EC2 Compute Unit
m2.2xlarge: $0.076 per hour per EC2 Compute Unit
m2.4xlarge: $0.076 per hour per EC2 Compute Unit
c1.medium: $0.034 per hour per EC2 Compute Unit
c1.xlarge: $0.034 per hour per EC2 Compute Unit

Amazon's high-memory on-demand instances end up costing slightly less than their standard instances if looking at the price per EC2 Compute Unit. However, at less than half the cost per EC2 Compute Unit are the high-CPU on-demand instances. These instances cost just $0.034 USD per hour compared to the high-memory instances at $0.076 or $0.085 for the normal instances. As our Linux benchmarks show from Amazon's compute cloud, when dealing with the properly multi-threaded tests the results are quite linear across the number of EC2 Compute Units.

The c1.medium instance though is not exactly great for compute-intensive purposes as it only provides five EC2 Compute Units and is limited to 32-bit software. However, that leaves the c1.xlarge, which packs 20 EC2 Compute Units, 7GB of RAM, and nearly 1.7TB of storage for just $0.68 USD per hour. While it has less compute power than the m2.4xlarge instance (20 vs. 26 units), it costs less than half of what the quadruple extra large memory instance will cost. With that difference, you could purchase two c1.xlarge instances and have 40 EC2 Compute Units while still coming out financially ahead than the m2.4xlarge.

If your needs are solely CPU focused, the sweet spot on Amazon's Elastic Compute Cloud is certainly with the extra large high-CPU instance (c1.xlarge). Though if your needs require a greater memory capacity, you will obviously need to stick with the high-memory instances.

With Phoronix Test Suite 3.0-Iveland and OpenBenchmarking.org (when available), if you are looking for results under a specific workload there are the results from all 60 tests that were carried out and can be directly compared to your own customized cloud instance or results running atop real hardware.