Does Distributed Power Management impose any requirements on the server hardware?
Generally speaking, most servers capable of running modern hypervisors like VMware ESXi and supporting automated resource distribution tools like Distributed Resource Scheduler (DRS) should also possess the basic requirements to support a power management method that is acceptable to manage a power-off (ACPI S5) state through Distributed Power Management (DPM).
One of the most popular methods of power management is wake-on-LAN (WOL), which uses control packets sent over the LAN from other clustered servers. The server's NIC, motherboard and BIOS must support WOL, and the NIC may need to auto-negotiate from 1 GigE down to a slower speed when powering down, meaning auto-negotiation should be enabled where possible. Ultimately, there are just too many network adapters, server models and BIOS versions to guarantee seamless WOL functionality on every possible combination of devices. IT architects should test WOL functionality before deploying the capability in production, or when upgrading servers and network components to ensure continued operation. This will help to prevent a new NIC or server from inadvertently "breaking" WOL operation.
Instead of WOL, DPM can also utilize out-of-band systems management technologies, including the Intelligent Platform Management Interface (IPMI) or dedicated hardware management platforms such as HP's Integrated Lights-Out (iLO). However, both of these methods rely on proper configuration of the server's baseboard management controller (BMC) to allow external power control. This is another case where suitable testing will help ensure that DPM performs as expected and delivers the anticipated benefits.
If the server is not currently capable of supporting ACPI S5 power control through DPM, it is still possible to use DRS independently without DPM by allowing the emptied server to use its own power-down mechanisms. For example, most modern servers will place idle processors in a C1 halt state -- saving power by not using the processor cores -- or use lower voltage and clock frequency settings to reduce processor power use, which is sometimes called processor throttling. In addition, the server may use a native "standby" mode (ACPI S3), which suspends the system to RAM. While all of these alternate methods can reduce the server's energy use, the system will still use more energy than if it were turned off. This deep savings is what makes DPM so appealing for busy virtualized environments.
Dig Deeper on DRS and DPM
Related Q&A from Stephen J. Bigelow
While the Windows Admin Center is one way to manage the Azure Stack HCI platform, you can also use traditional, battle-tested tools. Continue Reading
There are many tools available on the AWS Marketplace for QA testing, making it difficult to determine where to begin. What should an enterprise look... Continue Reading
Hyper-converged infrastructure that runs on Windows Server is not a new concept, but Microsoft's Azure Stack HCI program has one big difference from ... Continue Reading
Have a question for an expert?
Please add a title for your question
Get answers from a TechTarget expert on whatever's puzzling you.