Cloud Hosting Provider Hardware Benchmarks

While running some hardware benchmarks on my new Core i7 Apple MacBook Pro, I got the idea to run the same benchmarks across several Cloud hosting platforms to see how their hardware stacks up. The tool I used to perform the benchmarks is Geekbench from Primate Labs. Now, this benchmark was designed to accurately test processor and memory performance so these benchmarks do not take into account disk i/o, network i/o, etc. In addition to the performance results, we learn a bit about the underlying hardware on the host machines that run these platforms. The different providers benchmarked here are: Amazon EC2, GoGrid, Linode, Rackspace Cloud Servers and Storm on Demand (by LiquidWeb).

Before we get to the results let me disclose the fact that I WORK FOR RACKSPACE (that is me on the front left). That being said, you can trust my benchmarks or not — that is up to you. Believe me when I say that I would not risk my credibility and reputation over some simple benchmarks. The good part is that you can very easily replicate these results on your own in just a few minutes and for less than a few dollars (total across all providers — isn’t the Cloud great?) if you want to see results first hand.

All providers were kicked with Debian 5.0 (64 bit) except for GoGrid where I had to use CentOS 5.3 (64 bit). The great thing about Geekbench is that there are no options, so no confusion. You can run the tests in 32 bit or 64 bit mode depending on whether or not you have a license. These tests were run in 64 bit mode so I could test the full potential of these systems.

Provider Amazon EC2 GoGrid Cloud Servers Linode Rackspace Cloud Servers Storm on Demand Servers
Server Size Large 2 GB 2080 2 GB 2 GB
Integer Score 1561 3183 3105 3979 2513
Floating Point Score 2438 4293 3919 5304 3568
Memory Score 1839 3431 2696 2485 5374
Stream Score 1375 4355 1991 1432 4945
Overall Geekbench Score 1904 3738 3196 3889 3697
Platform Linux x86 (64-bit) Linux x86 (64-bit) Linux x86 (64-bit) Linux x86 (64-bit) Linux x86 (64-bit)
Processor Dual Core AMD Opteron(tm) Processor 270 Intel(R) Xeon(R) CPU           E5520  @ 2.27GHz Intel(R) Xeon(R) CPU           L5420  @ 2.50GHz Quad-Core AMD Opteron(tm) Processor 2374 HE Intel(R) Core(TM) i5 CPU         750  @ 2.67GHz
Processor ID AuthenticAMD Family 15 Model 33 Stepping 2 GenuineIntel Family 6 Model 26 Stepping 5 GenuineIntel Family 6 Model 23 Stepping 6 AuthenticAMD Family 16 Model 4 Stepping 2 GenuineIntel Family 6 Model 30 Stepping 5
Logical Processors 2 2 4 4 1
Physical Processors 2 1 1 1 1
Processor Frequency 2.00 GHz 2.27 GHz 2.50 GHz 2.20 GHz 2.67 GHz
Memory 7.51 GB 1.96 GB 2.81 GB 2.01 GB 1.68 GB
Processor Cores 2 2 4 4 1

*Note: You can click on the provider name at the top for a direct link to the Geekbench results browser.

You will notice that I tried to keep all the test instances right about the same size, or as close as I could. The only big variation comes with Amazon where the smallest instance I could build was of the Large variety (thought that might have given it an edge but apparently not). The Linode instance is positioned to be a bit more powerful as well and offers almost a full 1 GB more than GoGrid, Rackspace, and Storm due to their funky memory/price scale.

Now, we all know and understand that resources on the host machines that power these instances can fluctuate depending on many factors such as the time of day and extra resources that may be available (CPU bursting for example). Because of this I ran each benchmark 3 different times throughout the day: early morning, mid-afternoon, and late evening (~11:00PM CST). Every single provider showed their best results during the late evening test and these are the results posted here.

If you have any questions or feedback please leave a comment and I’ll do my best to respond as quickly as possible. Thanks for reading!

  • Pingback: Tweets that mention Cloud Hosting Provider Hardware Benchmarks (Rackspace, Amazon, Linode, Storm, GoGrid) | Chad Keck -- Topsy.com()

  • Anonymous Coward

    Thanks for posting this up. Because of the contention multiple cloud servers can create on a storage device, I’d be interested to see what kind of numbers you get there if you ever decide to run tests on that realm of performance. My guess is the reliability of the numbers would be hard to establish because of the various I/O requirements of other cloud servers shared on the same hardware as your cloud server… even still I’m curious. :-) Thx again

  • Robin

    Doesn’t Joyent merit benchmarks anymore?

    • Who? Haha, I’m just kidding! Joyent isn’t part of this benchmark because their products are more platform based (like Rackspace Cloud Sites). I’ll be doing some benchmarks of PaaS type systems in the near future and will be sure to include Joyent.

  • So, highlighting the best numbers in green, or blue, and highlighting the worst numbers in re, or orange, would make the chart significantly easier to peruse quickly.

    kthxbai

    • Yeah, I know…but the WordPress editor is giving me a hell of a time with these tables =\

  • jsherk

    So clearly, amazon is way at the bottom of the list!!

    But what do the numbers really mean? Storm beat out Rackspace on Memory and Stream … so what kinds of applications would that relate to that might benefit from these higher Memory and Stream scores?

    And conversly, what kinds of applications would benefit from the higher Rackspace scores on Integer and Floating point?

    What about VoIP phone service … which scores might be most relevant to these kinds of applications?

    Thanks for review.

    • I’m looking for someone much smarter than myself to answer those questions, haha!

      Actually, I intend to do some more real world benchmarks against Apache (as I’ve done before), MySQL, etc in the near future.

  • Greg

    Chad, thanks for posting this. I was just thinking about doing something similar this morning because I’m thinking about moving our app to the cloud.

    Do you have a theory as to why Storm does so much better on the Memory and Stream tests?

    • Not a problem!

      No theory on why Storm is better on the memory and stream tests other than maybe the memory bus is faster on that particular Intel chipset. Not sure what real world impact this would have versus slower Integer and Floating Point scores on the processor. Not the hardware geek I used to be a long time ago :)

  • ChrisGaun

    Chad, I was wondering does GeekBench automatically tell you what processor the instance is running on?

    Maybe you don’t remember since this post is old but it would help me a lot.

    Best
    @Chris_Gaun
    chrisg@ideasinternational.com

    • Hey Chris,

      Yes it certainly does. GeekBench by default will return all the processor details including the model, processor ID, frequency and more.

  • Nice write up Chad, especially the comparison put in the tabular format. It helps with making the comparison on the fly.
    Thanks.
    Regards,
    Stan from Cloud Hosting

    • Glad it was helpful. Time for a refresh of the data though I think. Will probably do this in the next few weeks.

      –Chad

%d bloggers like this: