Remember when, if you heard someone say, “I can see right through you,” it was clear you weren’t getting a compliment? Today, however, it seems that everyone wants transparency—especially when it comes to energy usage.
According to a recent article in Data Center Dynamics, the biggest obstacle thwarting datacenters and cloud companies from realizing energy transparency is encapsulated in the story’s subhead, “We can’t fix what we don’t understand.”
The article cites the work of Dr. Aman Shehabi, research scientist at Lawrence Berkeley National Lab, who has spent more than a decade striving to better understand energy usage despite rapid market changes and pervasive industry secrecy.
We realize that some datacenters and cloud companies might prefer to be purposely vague or secretive about the extent of their power usage. Perhaps it’s for confidentiality or—worse yet—to camouflage the true enormity of their appetite. Still, it seems like datacenters often operate like they’re playing an energy-related version of “Monopoly megawatts.”
The simple reality could be that datacenter operators have suffered from a lack of available methods for determining how much power they actually use. Therefore, they haven’t grasped the full extent or results of their power consumption problems.
Luckily, technology advancements, such as the emergence of Software Defined Power (SDP), enable datacenters and cloud companies to achieve much-needed power-usage transparency. The result is the ability to create real-time power profiles. What’s revealed are existing constraints and pinpointing of the impact of power spikes at certain times of the day, as well as different times of the year depending on the company, business model or external events.
For example, after capturing data from a well-known CRM company, SDP showed power spikes typically on Mondays and Tuesdays when everyone returned to work and accessed all their apps. For an online retailer, dramatic spikes were easily correlated to specific holidays or sales events that produce massive sales volumes.
Aside from getting a firm fix on exact energy usage, power profiles measure many variables, including voltage, current and wattage. Then with the aid of AI and machine learning, all these data points are analyzed over a timeframe— from as small of a sample as 7 days to a larger set of data points collected over years” – to predict future consumption.
This valuable insight is essential to developing what we call “power awareness,” which is an essential step to determining viable energy-saving solutions. In some cases, increasing energy utilization can be achieved by improving workload orchestration. This is accomplished by migrating power-hungry applications to an alternative site during peak processing hours. In other cases, the answer is “peak shaving,” where power capacity is extended by tapping into unused capacity that’s been assigned to handle peaks and buffers, along with redundant and back-up power.
Everyone involved in a datacenter and day-to-day operations has a stake—and a voice—in building a power-aware profile of energy usage. Traditionally, these stakeholders only gathered aggregate energy and other operational information. The absence of accurate details on a more intricate level, such as per-rack, -row or –server, is what led to the long-held view that oversubscribing and overprovisioning power was the only way to ensure sufficient capacity to accommodate energy spikes and peaks.
Thanks to SDP and the ability to gather granular stats about usage, this antiquated and costly power provisioning practice no longer is needed. The ability to create power profiles with up-to-date awareness of energy utilization permits fine-tuning application performance and on-the-fly changes that result in more reliable, efficient and well-balanced datacenters. The fix is in: Software Defined Power can play a starring role in achieving the kind of transparency everyone desires.