The Effects of HP's Power Profile on vCPU Performance

When you buy a new HP server it comes with the BIOS setting HP Power Profile set to Balanced Power and Performance. That sounds good but is bad - at least for CPU performance in a VDI environment.

When we prepared our servers for use as virtualization hosts for Citrix XenDesktop we took great care to configure the HP Power Profile for Maximum Performance:

BIOS setting - HP power profile

That fixed the problems with really awful CPU performance we had in the beginning. But still - some VMs were performing worse than others. When I looked at the problem I soon found a VM with bad CPU performance. I ran some benchmarks, comparing it to a “good” VM:

Test setup:
CPU: Xeon E5-2670 2.6 GHz
Hypervisor: VMware ESXi 5.0 Update 1
VM OS: Windows 7 64 Bit
VM configuration: 2 vCPUs

BenchmarkGood VMBad VM
Cinebench 11.5 CPU » more is better2.201.48
NovaBench CPU Test » more is better262204
Super Pi 1M [s] « less is better13.232.2

Depending on the benchmark, the “good” VM is between 28% and 244% faster than the “bad” VM. That is a lot!

I suspected the power profile setting of the server hosting the “bad” VM to be wrong, but how to check without rebooting the server and accessing the BIOS? These are production systems and we cannot use vMotion to move the VMs off the server because the VMs are on local storage. Luckily the power profile setting is not only available from the BIOS, but from the iLO console, too. It can even be changed from iLO, taking effect immediately:

iLO Power Settings

Sure enough, the “bad” VM was on a server with incorrectly configured power profile. Switching it to HP Static High Performance Mode fixed the issue. Strangely, the settings are named differently in BIOS and iLO.

If you want to know more about power profiles and power regulator: HP has a whitepaper on the topic.

CitrixVCPUVMwareXenDesktop

Comments

Related Posts

How-to: XenApp/RDS Sizing and Capacity Planning with uberAgent for Splunk

How-to: XenApp/RDS Sizing and Capacity Planning with uberAgent for Splunk
Do you know the maximum number of users each of your terminal servers can host with acceptable performance? You may have found out the hard way how many are too many - but how many are just right? Farm sizing and server capacity planning are typical tasks for consultants who often have a hard time fighting the peculiarities of perfmon and logman trying to get the data they need for their calculations. It can be so much easier at no additional cost. The 60-day Enterprise Trial version of Splunk in conjunction with an evaluation license of uberAgent give all the information you need in much less time. Here is how.
Citrix/Terminal Services/Remote Desktop Services

Latest Posts

Fast & Silent 5 Watt PC: Minimizing Idle Power Usage

Fast & Silent 5 Watt PC: Minimizing Idle Power Usage
This micro-series explains how to turn the Lenovo ThinkCentre M90t Gen 6 into a smart workstation that consumes only 5 Watts when idle but reaches top Cinebench scores while staying almost imperceptibly silent. In the first post, I showed how to silence the machine by replacing and adding to Lenovo’s CPU cooler. In this second post, I’m listing the exact configuration that achieves the lofty goal of combining minimal idle power consumption with top Cinebench scores.
Hardware

Fast & Silent 5 Watt PC: Lenovo ThinkCentre M90t Modding

Fast & Silent 5 Watt PC: Lenovo ThinkCentre M90t Modding
This micro-series explains how to turn the Lenovo ThinkCentre M90t Gen 6 into a smart workstation that consumes only 5 Watts when idle but reaches top Cinebench scores while staying almost imperceptibly silent. In this first post, I’m showing how to silence the machine by replacing and adding to Lenovo’s CPU cooler. In a second post, I’m listing the exact configuration that achieves the lofty goal of combining minimal idle power consumption with top Cinebench scores.
Hardware