The Industry

EC2 CPU Performance is Suspect February 16, 2008

Like many, I’ve been sort of infatuated with Amazon Web Services lately. My new projects almost always use S3 for backend file storage at this point; it’s not just cheaper, it’s also a simple way to segregate static files from your app servers. Overall, S3 works pretty much as advertised.

EC2, however, does not work as advertised, at least from a CPU perspective.

EC2 instances come in small, medium, and large. The cost of running a “small”, or even a “large, production server, full-time, is about comparable to that of running a traditional, entry-level dedicated server. Because it’s priced per-hour, and there are no setup or termination fees, it’s a bargain.

Here’s the catch. Although Amazon advertises the EC2 processor on a small instance as “equivalent” to a “dedicated 1.7GHz Xeon”, that’s really not how it shakes out. For some reason, which I hope more expert sysadmins will eventually determine, processes on an EC2 instance will never perform at that level—that is, it’s impossible to use even 50% of the available CPU time.

It’s a hot topic in the AWS forums:
CPU performance 50% of expected
CPU Utilization below 50%

Unfortunately, discussion of these issues invariably gets bogged down in super-technical discussion of benchmarking and hardware virtualization, so the outrage is a bit slow in building.

Clear is this: if you are running a CPU-intensive application, investigate and test thoroughly before assuming that EC2 is viable for your purposes. Yes, you get loads of RAM, and a complete virtual server, which is kickass… but the CPU may significantly underwhelm you. Check it out.