Performance encompasses a lot of different metrics. In this example there are only a handfull that matter to me: CPU horsepower, Storage I/O, and Network Latency. In this post we’ll start with CPU.
There’s a great whitepaper benchmarking and comparing several aspects of the two platforms available via CloudCommons that touches on several points related to these metrics, so we’ll use that as a base.
One aspect of the two cloud environments that is missing from the benchmark series is the CPU power involved.
When purchasing EC2 compute time the different instance types not only have different memory and HDD allocations, but different CPU capabilities as well - what Amazon calls a Compute Unit. Each type has a set cap on CPU resources and if you need more power you have to spin up a new larger instance.
By comparison, Rackspace offers CPU Bursting. Like EC2, each instance size has a guaranteed amount of CPU resources. Unlike EC2 if there are additional resources free on the host node your VPS is free to utilize them. The ability for your workload to seamlessly utilize all the horsepower of that quad-core server processor while undertaking a particularly intensive task (like compiling a program from source) can be a huge boost.
The best example of this difference shows up in the obvious place, the UBench CPU test, but is also reflected in several other tests that tax the CPU more than other resources (7zip is another good example). In both tests performance on Rackspace is equal across all node sizes while the performance on EC2 scales up and down based on the instance type, indicating that spare cycles are apparently readily available on Cloud Server host nodes.
Another set of benchmarks by The Bit Source concludes that CPU Bursting could mean your Rackspace server is twice as powerful as a similar EC2 instance, though Rackspace does not offer customized high-CPU instances that could provide an exponential boost (at a similar cost increase) to performance.