Megatrends such as Cloud Computing, Big Data, Mobile and IoT have resulted in rapid expansion and growth for the large Data Center and Co-Location providers, as well as the Web 2.0 companies that have taken the Owned and Operated approach. Hyper-Scale customers have also pushed hardware vendors to optimize their solutions for space density, and while power consumption requirements have improved, the increased number of machines per rack means that the strategy for power distribution needs to change.
Virtualization and Hyper-Convergence have also contributed to more variance in workloads and underlying power-consumption, which has resulted in Data Center operators having to over-provision power in order to handle the associated spikes in power consumption needs.
We have applied “software-defined” controls to the rest of the Infrastructure, so why not take this innovative approach to the management of power distribution? Imagine if you could unlock currently unusable capacity and achieve an increase in workload performance of 20- 50%? First, let’s step back and look at some other trends and potential solutions for data center power management.
Metered Power Strips - The concept is simple, if you want to understand your power consumption and distribution, you need to measure and track it. This solution gives you that ability, but needs to be part of a larger overall platform to correlate all of the data.
208V vs. 120V - Amps * Volts == Watts, so by increasing the Volts from 120 to 208, less Amperage is required to power the equipment. Below is a chart (Credit: HP) that shows the advantages.
Power Usage Effectiveness (PUE) - This is also a formula. An “Ideal” PUE rating is 1.0, but the metrics to achieve that can vary wildly, and there are several articles around the fact that PUE was simply “Green Tech Hype”. I will leave my position as this is a challenging metric to quantify, so it’s something important to be aware of, but is not the be-all, end-all metric.
UPS/Battery Monitoring - As with metered power strips, this solution is only as effective as it’s integration with an overall DCIM platform. The other challenge is that it’s a defensive approach, instead of trying to play offense and unlock power sources that are either inefficient or underutilized.
Throttling/Capping Servers - Managing the power consumption of individual servers is challenging. At best, it is a coarse-grained approach lacking orchestration. More importantly, capping CPU power severely degrades the workload performance resulting in frequently dropped requests. Service levels suffer and in a highly competitive cloud provider landscape this is not viable. For colocation vendors, this typically isn’t even an option since they cannot interfere with the IT equipment as it is their client’s property.
What is needed is an overall platform that ties together the best solutions and applies “software-defined” technology to fully control every aspect of Data Center Power Distribution and Monitoring. I will write more about this in a future blog post. Stay tuned…