There are a number of technologies available to help reduce power consumption in modern data centers, but how much should we take advantage of these options? Let’s focus on a few technologies than can really help reduce power consumption (and reduce thermal output) in the data center, but most importantly, what do you think about how beneficial they will be?

The first technology I will focus on is related to storage systems — disk spin down. This effectively will power down idle hard drives in an array, with the option to resume their activity either by policy or activity. A number of products may also call this a MAID (massive array of idle disks) or intelligent power management (IPM). One of the storage professionals I look to for good information, Ray Lucchesi has some notes on disk spin down on a popular storage product. This of course is for a tier of storage that can absorb this type of latency, but I’m not exactly sold on this approach for most storage systems. I think that this type of power consumption strategy would only work for very large storage systems which can absorb high latency in idle areas or backup data tiers. This would be a tough sell for traditional primary storage systems.

Another technology that I really like is VMware’s Distributed Power Management (DPM). DPM works for VMware DRS clusters and allows virtual machines to be consolidated to a smaller number of hosts, and the idle hosts sent to a standby state. Take an example cluster of five ESXi hosts, and should the active workload be consolidated on one, two or three fewer hosts, the vCenter Server can use vMotion migration to evacuate the hosts. Then, the hosts would be sent into a standby state. The host is resumed via a wake-on-LAN packet from another host at the time in which the workload increases again. Remaining hosts in the cluster are then capable of sending the wake-on-LAN packet to bring one or more hosts out of standby. In my experience this feature is very good in that an entire host is idled, and resumes when the workload dictates it. What I don’t like about this feature is that there can be a monitoring void for solutions that are out of band. Specifically, any paging system would likely determine that the host is down; yet, instead it was sent to a standby state.

For individual physical servers, traditional idling or power saving isn’t a viable option. Managed systems, such as a storage controller in the disk system example and VMware vCenter in the DPM example — in my opinion — are the only ways to go to aggressively reduce power consumption in the data center.

What do you do for aggressive power reduction? Do these ideas appeal to you? Share your comments below.