How Not To Save Energy with Windows Server 2008
Having read Microsoft’s documentation on how Energy Efficient Windows Server 2008 is compared to Windows Server 2003, I thought I’d show my findings with some Real World Testing. In the documentation found here (yes, you can only read it if you have Microsoft Office 2007), the graph below is presented:
The following footnote is also included:
Power comparison of server unit only. Does not include external drive arrays.
This immediately made me realise that the results of the graph would be inaccurate as this means that they will have not installed the Operating System onto the local hardware but a SAN-type array. Meaning that the power usage for the discs would not be included. Anyone who has used Windows Server 2008 or indeed Windows Vista will know that there is excessive disk usage even when the computer is seemingly idle.
The hardware I used for this test is a Hewlett Packard DL380 G5 with the following specifications (taken from the Lights Out Management Card):
Processors (x2)
| Processor Speed | 2333 MHz |
| Execution technology: | 4/4 cores; 4 threads |
| Memory technology: | 64-bit capable |
| Processor 1 Internal L1 Cache: | 128 KB |
| Processor 1 Internal L2 Cache: | 8192 KB |
Memory
|
DIMM 1A : |
1024 MB 667 MHz |
|
DIMM 2C : |
2048 MB 667 MHz |
|
DIMM 3A : |
1024 MB 667 MHz |
|
DIMM 4C : |
2048 MB 667 MHz |
|
DIMM 5B : |
1024 MB 667 MHz |
|
DIMM 6D : |
2048 MB 667 MHz |
|
DIMM 7B : |
1024 MB 667 MHz |
|
DIMM 8D : |
2048 MB 667 MHz |
8 x 15,000RPM Small Form Factor (SFF) Serial SCSI (SAS) 72GB Drives
The server also has Dual Fully Redundant Power Supplies and Fully Redundant Fans. More information about the DL380 G5 Series can be found here.
I then proceeded to put the server under a few tests and I was surprised with the results. I had a feeling that the Microsoft White Paper would be inaccurate, but I didn’t realise exactly have inaccurate it would actually be. Using HP’s Integrated Lights Out Version 2, I was able to get the Power Usage Graphs from the servers for the last 24 hours. Neither Operating System was given any different Power Management Settings to the defaults. The results are below:
Test 1: Microsoft Server 2003 R2 (Idle)
Test 2: Windows Server 2003 R2 (running Microsoft Virtual Server 2005 R2 with 7 Virtual Machines):
Test 3: Windows Server 2008 (Idle)
The graphs show that over the 24 hours, on average, Windows Server 2008 was using 43.8% more power than Windows Server 2003 in an idle state. It also shows that Windows Server 2008 uses 26.5% more power than Windows Server 2003 whilst under normal load.
We haven’t been able to test Windows Server 2008 under normal load thus far, as we’re struggling to get anything that will play nicely on Windows Server 2008 including Microsoft’s own Hyper-V and System Centre Virtual Machine Manager 2008.
