Posted by: CarlBrooks
cloud benchmarks, cloud metrics, sick man in a sack race
“Benchmarks!” you explode, your exquisitely sensitive logical faculties coruscating in outrage. “Benchmarks are vendor-driven popularity contests that cherrypick tests for results! Its a swamp of dismal nonsense, a perpetual statistical hellhole that means nothing! Benchmarks make engineers maaaad!!” you say.
Well, fine, then you do it, because there aren’t any for cloud yet, and I don’t care if Rackspace did go have it done. They had sound reasons, there is precious little precedent, and the results are informative, useful and not overtly flawed.
Analyst and high-traffic expert Matthew Sacks carried out the benchmarks on his website The Bitsource. Overall, the tests were rudimentary and carefully controlled. He tested the time it took to compile a linux kernel on every type of instance between the two services, and used Iozone to compile read/writes for storage systems.
Sacks, a systems administrator at Edmunds.com, said he developed the methodology himself.
“The idea behind it is that we can get a pretty good idea on how these instances stack up” he said, but it’s not a comprehensive metric. “I decided on kernel compilation as a general measure of CPU” he said, because it was well understood, easy and fast to replicate and uncomplicated in results.
Sacks said that Rackspace’s motivation was good old fashioned boosterism. “They had received reports from their customers that Rackspace was way better than EC2.” he said, so they decided to test that out with a third party. The results seem to bear them out.
“There are clear wins in CPU and disk performance,” he said. Rackspace beat EC2 instances in compiling by a slender margin in every case but one, and showed 5 to 10 times the amount of CPU availability. Disk Read/Write was also higher, sometimes by twice as fast, although random access tests were much closer, suggesting throughput on EC2 lags behind Rackspace even if data request execution doesn’t.
However, Sacks took pains to say that his tests did not mean that an application running on EC2 could be shifted to Rackspace and save either time or money by default. He said that users always had to consider the application, not just the infrastructure, and he wanted his tests to be a resource for people to come and compare their specific needs. “The variables are so great it’s hard to come up with a standard for testing [cloud]” he said.
“What I would like to do is test more providers,” said Sacks. Perhaps he’ll get the opportunity- he said the testing and review only took a few weeks and it wouldn’t be hard to repeat for different platforms.
Sacks’ two-man experiment aside, I can think of at least one cloud out there that could make EC2 and Rackspace look like a snail chasing molasses when it comes to kernel compilation and disk I/O:
“NewServer’s Bare Metal Cloud makes Cloud Servers and EC2 look like two sick men in a sack race” sounds catchy to me. I wonder if Rackspace will sponsor that benchmark?
But that’s the point– I want more third party tests like Sacks’. I want more and more and more independent review of what providers say and what they do. I can find more compelling independent information about the USB stick in my pocket than any one of the clouds that want me to trust my business to them.
It only took Matthew Sacks a few weeks to make a clean, well documented and useful set of benchmarks, even if that came at the behest of a vendor.
So let’s make more.