Amazon EC2 Cloud Benchmarks
Phoronix: Amazon EC2 Cloud Benchmarks
Following last week's KVM vs. VirtualBox benchmarks and then looking at the multi-core scaling of KVM virtualization, we now have up some benchmarks of Amazon's EC2 (Elastic Compute Cloud) looking at the performance of this leading cloud computing service. This is just the start of some Amazon EC2 benchmarks with this article looking at the performance of their m1.large and m1.xlarge instances compared to some other hardware. There is also an OpenBenchmarking.org ID for those interested in replicating these tests.
Very good idea Michael! Thanks for this benchmark!
What about Rackspace?
I'd love to see Amazon EC2 compared not only to real hardware but also to Rackspace's cloud offerings. Some people say they fare a lot better in terms of CPU than EC2.
Censored service, not gonna use, plenty of alternatives.
Was just about to post:
Originally Posted by mirza
Amazon... Wikileaks... 'nuff said
It might be interesting from a technical viewpoint but:
a) NEVER trust your data to a cloud
b) especially not Amazon (-> Wikileaks)
So this should be a big no go area for sensible people.
I don't know about that. There are a few cloud video encoding services that do the job quite well. They are extremely handy in situations where you may have something like 50+ personalized training videos a day to render day and get it to their clients within hours.
Originally Posted by popper
So it's "similar" that the Opteron was between the m1.large and m1.xlarge in the previous benchmark, but in this one, the Opteron is between 2 and 3 times faster than either of the AMIs? Think you need to watch how much gluhwein you're drinking, Michael...
Originally Posted by phoronix
Close? The fastest (the Core i7) is well over twice as fast as the slowest (m1.large). How is that close? I guess you could say the xlarge and the Opteron are sort of close, but that's not the only conclusion to draw here, and you certainly didn't specify that.
Originally Posted by phoronix
Article nitpicking aside, I think it would be funny to write a program that automatically looks at graphs like this, draws "conclusions", and formats them in a style that resembles Michael's writing I have some spare time this holiday season, so I might take a stab at writing such a program! It will be hard to exactly mimic his writing style, though... he has have had had had had had had had had -- er, I mean, has a unique style
hard to test
Cloud services seem to be a flame generating topic, but I'll try to stay above the fray..
The trouble with benchmarking AWS is that there are so many options. Let me provide an anecdote: One of our instances was having terrible DB performance and we were about to abandon EC2 until we brought a couple of ELB disks online and put the DB on a RAID. By paying slightly more, we were able to increase performance from unacceptable to better than a standalone server at less than half the cost.
There are just too many variables in play. On a standalone, you have to support the hardware, pay for electricity, etc. On a cloud, you pay incrementally for everything, and resources are shared so you have to be able to scale your computing in ways that work with the cloud infrastructure.
One word on Rackspace cloud: It is a competitive alternative, but provides fewer choices than AWS. If you fit exactly into one of the Rackspace Cloud profiles, it is definitely the better deal, but if not, you're out of luck. It's not like AWS where if you need more memory, you choose the more memory option -- with Rackspace you have to choose the more cpu, more disk, more memory option. Again though, if your app needs all 3, then it is the better deal.