CentOS 5.5 Results on EC2
I'm having a problem with the Linux distribution of Geekbench running on CentOS 5.5 (using either Rightscale or Scalr EBS AMIs) on EC2 m2 instances. The CPU allocation for each of the m2 instances is as follows:
m2.xlarge: 2 cores
m2.2xlarge: 4 cores
m2.4xlarge: 8 cores
However, I'm getting roughly the same Geekbench score for each of these. Here are examples:
m2.xlarge (2 X5550 cores): 3377
http://browse.geekbench.ca/geekbench2/view?id=382689
m2.2xlarge (4 X5550 cores): 3534
http://browse.geekbench.ca/geekbench2/view?id=383503
m2.4xlarge (8 X5550 cores): 3537
http://browse.geekbench.ca/geekbench2/view?id=364405
Do you know what the problem could be here?
Keyboard shortcuts
Generic
? | Show this help |
---|---|
ESC | Blurs the current field |
Comment Form
r | Focus the comment reply box |
---|---|
^ + ↩ | Submit the comment |
You can use Command ⌘
instead of Control ^
on Mac
Support Staff 1 Posted by John on 24 May, 2011 02:55 AM
Hi Jason,
Thanks for your message. I looked at your Geekbench results and it looks like while EC2 is reporting that 2, 4, or 8 cores are available, EC2 is only providing 2 cores to your instances. If you examine the multi-threaded workloads (which are the workloads that take advantage of multiple cores), you'll see that in the best case they're only 2x faster than the single-threaded workloads. This generally indicates that only two cores are available.
Let me know if you have any other questions and I'd be happy to answer them!
Best,
John
2 Posted by Jason Read on 24 May, 2011 04:42 AM
John,
Could it be a problem with how the OS is working with Geekbench, because using a non-RHEL AMI such as Ubuntu, I'm able to get about triple the score on the same m2.4xlarge instance size.
Jason
3 Posted by Jason Read on 24 May, 2011 07:06 AM
Here are tests we ran about a year ago on the same instance sizes, but with a slightly older CentOS AMI:
m2.xlarge (2 X5550 cores): 5877
http://browse.geekbench.ca/geekbench2/view?id=241462
m2.2xlarge (4 X5550 cores): 5163
http://browse.geekbench.ca/geekbench2/view?id=241409
m2.4xlarge (8 X5550 cores): 4049
http://browse.geekbench.ca/geekbench2/view?id=241403
Our Unixbench parallel results do show a linear increase in CPU performance with the different m2 instance sizes on the same instances where Geekbench shows flat results, so it seems like the CPU resources are available, but Geekbench is not incorporating them into the results for some reason.
4 Posted by Jason Read on 24 May, 2011 05:31 PM
Here are our Unixbench results running on the same instances where we are seeing the flat Geekbench results:
m2.xlarge (2 X5550 cores): 929
http://benchmarks.cloudharmony.com/unixbench/0311-ec2-use-ec2-us-ea...
m2.2xlarge (4 X5550 cores): 1559
http://benchmarks.cloudharmony.com/unixbench/0311-ec2-use-ec2-us-ea...
m2.4xlarge (8 X5550 cores): 2396
http://benchmarks.cloudharmony.com/unixbench/0211-sc-ec2-us-east.li...
So, Unixbench is clearly getting access to and utilizing the additional cores on these instances. Any idea why this would not be the case for Geekbench?
Support Staff 5 Posted by John on 04 Jun, 2011 12:42 AM
Hi Jason,
Interesting! Thanks for the links. We'll dig into it and see if there's a problem with Geekbench that needs to be addressed.
Thanks,
John