Unsolved
This post is more than 5 years old
1 Rookie
•
6 Posts
0
21416
July 27th, 2013 17:00
Performance Mode on R620 increases Application Performance 235% ! Why?
We have a new Dell R620, 48GB , Dual E5-2630, H710 Controller with 2 600GB SAS 10K in Raid 1 under Windows Server 2012. This was intended to replace a much older Dell Edge PowerEdge with Dual 1.6ghz E5310's 20GB with 7200RPM SAS on Perc 5i's running Windows Server 2008 R2.
We are using this as an application server.
To our great disappointment - after setting up our app and testing the applications performance - our new box ran 20% slower than the old one.
After reading various posts trying to discover why - we saw a suggestion to try changing the windows power settings (didn't improve anything) and then to try changing the system performance setting in the Bios from the default (I think it is Performance Per Watt) to Performance.
After changing the the system to the Performance setting, we noticed immediate improvement - almost 20% on the Passmark CPU/Memory Benchmarks.
What was surprising was a 235% improvement in Application Processing performance in our intensive application processing - which was a great relief.
So while we are now much happier with our new Server, we wonder what on earth would cause the server to perform so much differently? Anything to do with Windows Server 2012?
Should the 12th Generation Servers come with a Black Box Warning ("Application Processes may run very slowly with Default Bios Settings")?
Daniel My
12 Elder
•
6.2K Posts
0
July 27th, 2013 17:00
Hello Glenn
The performance setting disables several power saving features. The CPU and memory run at higher voltages, and do not downclock during idle times. Your benchmarks may be a little skewed as there is a brief amount of time required to increase speeds on the CPU and memory from idle. This can cause benchmarks to report inaccurately if they are short tests.
You can read more about it in this white paper: http://en.community.dell.com/techcenter/extras/m/white_papers/20248740.aspx
Thanks
Glenn Barber
1 Rookie
•
6 Posts
0
July 27th, 2013 20:00
While the widely used PassMark Server benchmarks run about 10 seconds for each of 10 tests, our application benchmarks are real world intensive processing involving disk/sql access, memory and cpu operations and are 4-5 minutes each in length. I would assume that while this is only a single user testing, they are long enough (and repeated run) to gather a realistic appraisal of the performance changes as a result of the bios changes.
Your response seems to suggest a disbelief in our observations - as I did myself until we had repeated our testing.
Glenn Barber
1 Rookie
•
6 Posts
0
July 29th, 2013 11:00
I might add - that when the machine was configured with the Performance Per Watt and we ran our intensive test - none of the CPU cores recorded more than about 30% - it was almost as if the machine was still in snooze mode. When we shifted to performance mode, some of the cores jumped past 80%.