The Dirty Secrets of Server Power

Power efficiency remains a top decision point for many customers looking to modernize their data center. Customers searching for an accurate way to compare server power among vendors are often directed to use power calculators without taking into account the many factors that drive power utilization. This notion preys on the simplistic nature of the power beast – lower number is better… but are we missing the (power) bus entirely?

The dirty little secret is that there is no industry standard for power calculators, and vendors can essentially publish a calculator around whatever workload they like (or no actual workload at all). Cisco UCS relies on the traditional spirit of the power calculator, using it as an engineering tool to estimate maximum power and  prevent customers from overloading electrical circuits. Other vendors choose to create liberal power calculators, encouraging customers to use it as a tool to compare across vendors. Needless to say, production environments vary wildly, and no single test characterizes all workloads or the frequency with which they occur.

https://www.youtube.com/watch?v=lvcLn7kD8-8

Tired of misinformation, Cisco performed exhaustive power efficiency tests comparing UCS to HP BladeSystem, IBM Flex System and Dell PowerEdge enclosures running SPECpower_ssj2008, an industry standard benchmark that measures efficiency (performance-per-Watt). No gimmicks, no trickery, no hidden calculators, open kimono. It comes as no surprise that Cisco UCS has better power efficiency as the domain size grows and UCS Fabric Interconnect power is amortized over more blades. In fact, the data shows that UCS is within a light bulb or two vs.  legacy solutions with very few blades. As the blade count increases, however, the power savings favor UCS and increase dramatically as more blades are added to the domain. For customers running high blade counts and large data centers, this equals real savings – potentially many Kilowatt savings. But what does that really mean for your environment?

The Hardware Power Distraction… Are you really chasing a fraction of a percentage point of energy efficiency based on worst-case projections? Is the power difference of plus or minus a light bulb for low blade counts really going to make a buying decision for those of you without large infrastructures? If power calculators are the dirty little secret, the dirty big secret is that true energy efficiency still comes from increasing system utilization. Microsoft’s IT Energy Efficiency Imperative¹ asserts that for every 100W of power into a data center, less than 3W are associated with actual computing; much of this waste due simply to underutilization. We see this every day where customers have legacy production environments they refuse to touch for fear of breaking something. Two things happen – they overprovision by 100% or more in the name of growth when they buy, or worse add new servers right next door for fear of upsetting the existing running environment. One day you look up and realize that you have various generations of systems in various states of their lifecycle running at 5% or 10% average overall utilization.

By developing UCS around Unified Fabric and policy-based management, Cisco has revolutionized customers’ ability to scale and right-size environments. With UCS, customers can literally double their blade resources in a matter of minutes, and policy-based computing with UCS Manager allows customers to take advantage of those resources accurately and without the headaches of setting up new switches for every 14 or 16 blades. With dynamic profiles and templates that adapt to any server and any generation in the domain, UCS truly enables customers to grow/shrink applications quickly while providing efficient workload mobility. This is the true power efficiency win in the data center and what continues to set UCS apart from the competition in the data center.

Share