Amazon EC2 Cloud Benchmarks Against Bare Metal Systems

After earlier this month delivering new Amazon EC2 cloud benchmarks, several Phoronix readers had requested some reference benchmark results against some "bare metal" hardware systems. For those not running their own performance comparisons against Amazon's Elastic Compute Cloud, here's some more benchmark results showing different EC2 cloud instance types compared to an Intel Ivy Bridge and Haswell system.

Today's data extends the Amazon EC2 performance benchmarks from earlier this month with results from an Intel Core i3 3217U "Ivy Bridge" and Intel Core i7 4750HQ "Haswell" systems. These are just two reference systems to give a better idea for how Amazon's m1.small, m1.medium, m1.large, m1.xlarge, and m3.xlarge EC2 instances compare to this modern Intel hardware with the benchmarks running on the systems themselves without any form of virtualization/container. These bare metal results were coming from an Ubuntu 13.10-based system, while the EC2 cloud was running Amazon's AMI, so besides the obvious differences, take the system

M1 Small - m1.small - 1 ECU / 1 Core / 1.7GB RAM / $0.060 per Hour

M1 Medium - m1.medium - 2 ECU / 1 Core / 3.7GB RAM / $0.120 per Hour

M1 Large - m1.large - 4 ECU / 2 Cores / 7.5GB RAM / $0.240 per Hour

M1 Extra Large - m1.xlarge - 8 ECUs / 4 Cores / 15GB RAM / $0.480 per Hour

M3 Extra Large - m3.xlarge - 13 ECUs / 4 Cores / 15GB RAM / $0.500 per Hour

These results are just some quick data points to give a better idea for rough expectations of the different Amazon cloud instances. Via the open-source Phoronix Test Suite software, you can easily repeat these benchmarks in a very easy, fully automated, and reproducible manner.

With the Phoronix Test Suite installed on any Linux/BSD/Solaris/OSX system, it's simply a matter of running phoronix-test-suite benchmark 1309198-SO-AMAZONCLO37 from the command-line to download all of the captured test data, download and install the selected tests and their configuration files, and then it proceeds to run the benchmarks and compare the results side-by-side via integration with OpenBenchmarking.org. There's also commercial support available and other custom engineering services for the dozens of enterprise-class features that our open-source benchmarking software has queued up over the half-decade of its public existence and nine years of internal development.