Administrator > Performance

The Mainframe Energy Efficiency Value: Part 2

A Contrary Opinion

During the course of my research, I was delighted to find a counter-opinion here. This article describes how x86 architecture can be twice to 15 times as energy efficient as mainframe architecture. The author argues that if an IT manager were to run approximately the same workload on an HP system (ProLiant DL580) versus an IBM z9, the HP system would outperform a mainframe in terms of energy efficiency. And this author may be right if you snapshot an HP system at any given time when it’s executing a workload, it may appear to be more energy efficient than a mainframe at that given time.

On the other hand, my company, Clabby Analytics, notes that every server in a distributed network doesn’t peak simultaneously (hence the above individual snapshot doesn’t represent performance characteristics of a distributed networked environment). Accordingly, when evaluating distributed system utilization, the sum of the interval data (data that shows what’s going on in a distributed computing environment from one period of time to the next) needs to be measured over time, not just in snapshot comparison. By measuring interval data over time, most enterprises will find that the composite peak of underutilized servers, back-up servers, failover servers, QA/test environment servers and the like will be very low—even lower than the 15 to 20 percent utilization figure mentioned in the first part of this series—even when production servers are operated at managed peaks of between 40 to 70 percent utilization. This means that although certain distributed servers at particular times may appear to be more energy efficient than mainframes, IT managers and administrators need to consider the bigger picture regarding how the entire distributed environment is behaving, not just a particular machine.

One outcome of considering the bigger picture may be that IT managers and administrators may find that when additional workloads are added to a mainframe, mainframes actually appear to perform better (because a mainframe doesn’t have to schedule additional servers and software to accomplish that workload. Instead, the mainframe resources are waiting in virtualized pools ready for assignment making the addition of new workloads essentially just an exercise in scheduling.

Summary Observations

Over the past year energy costs have fluctuated wildly across the globe and IT executives are now paying close attention to energy efficiency. This series has described several of the reasons why mainframe architecture is inherently more energy efficient than distributed systems architectures. Design, component efficiency and virtualization advantages all play major roles in mainframe energy efficiency.

But systems efficiency is only one piece of a large puzzle. What about storage energy efficiency? And beyond systems and storage, what about entire datacenter energy efficiency? Enterprises that run datacenters are already aware that energy costs related to supplying power (including the cost to operate uninterruptible power supplies) as well as the costs to cool/air condition datacenters can far exceed the cost to operate systems/storage environments. It’s important to move away from systems and storage devices that waste energy, but it’s of even greater importance to attack the macro problem of managing energy efficiently throughout an entire datacenter.