Big Iron--The Modern Mainframe: A Model of Space and Energy Efficiency

Newsletters	Subscriptions	Forums	Safari	Store	Career	Media Kit	About Us	Contact	Search	Home

Volume 4, Number 19 -- May 13, 2008

The Modern Mainframe: A Model of Space and Energy Efficiency

Published: May 13, 2008

by Chris Craddock

There is a lot of talk today about global warming, carbon footprints, and businesses "going green." In fact, in the opinion of many, energy consumption and environmental impact are among the most important issues presently facing society. These issues are particularly pertinent to IT for two reasons. First, IT consumes significant energy resources. IT organizations must therefore be fully engaged in any corporate effort to "go green." Second, IT organizations are under extraordinary pressure to do more with less. As energy costs rise, those costs consume a greater percentage of already-tight budgets.

IT organizations that are able to substantially reduce energy consumption (and, by extension, energy costs) will therefore free funds for allocation where they can deliver greater value to the business--such as research and development, acquisition of new technologies, and the building of additional capacity. So, above and beyond any sense of social responsibility, IT executives can aggressively embrace "green computing" out of sheer self-interest.

But what's the best way to go about "going green?" What strategic decisions can help control IT energy costs? And how can those decisions be defended against the various skeptics one is likely to encounter in the political process that is associated with all strategic decision-making?

Overcoming Resistance

Ohm's Law, discovered and published by the physicist Georg Ohm in 1827, describes the behavior of every electric circuit, from the simple ones that deliver power to your living room lights to the highly complex ones that make up a modern microprocessor. This simple equation--V=IR, where V is the voltage, I is the current, and R is the resistance of an electric circuit--is one of the most important equations in modern science.

You can think of these three variables by using a garden hose analogy. The conductor is the hose, and the electrons are the water flowing through the hose. The voltage is the water pressure. The current is the volume of water coming through the hose. And, as the name suggests, the resistance is like a kink in the hose--preventing the water from flowing freely from one end to the other.

If there were no resistance, electrons could rush around without losing any of their energy. In many places, it is advantageous to have electrons flow without minimal energy loss from resistance. This is the main reason that scientists are searching for superconducting materials.

However, with all of the typical electrical conductors we use today--mostly copper and aluminum--electrons always waste some of their energy in overcoming resistance. That lost energy is turned into heat in the conductor. This is why electrical equipment is almost always warm to the touch while it is running.

Modern microprocessors contain literally millions of tiny circuits densely packed together within an area not much larger than a thumbnail. Each circuit has some resistance and therefore throws off a tiny amount of heat. Multiply that tiny amount by millions and you get lots of heat from the microprocessor alone. But that is only a small part of the circuitry that makes up a computer system. It must be plugged into a mother board which in turn connects to a variety of memory and I/O boards. One or more fans pull cooling air into the interior of the system enclosure. All of this is powered by a stout transformer that reduces the input voltage by as much as 90 percent from the wall outlet and converts it to the low voltage DC current required by most of the internal components. So the transformer itself and the fan motor are significant sources of heat!

As much of that heat as possible has to be removed, or the circuitry will eventually fail. System fans pull cooler ambient air from outside the enclosure and blow it across the electronic components, cooling them by exchanging their heat with the air. The warm air then flows back out into the room. Air conditioning removes heat from the room air and the cycle continues. The electricity required for air conditioning can dwarf the amount of electricity required to power the systems themselves. As a result, enterprise data centers can wind up generating significant energy consumption and high energy bills.

This is the "double cost" of electronics. We need energy to power our electronic devices, and then we need more power to keep them cool.

Power Consumption On Distributed Servers

To understand how the rudimentary principles of electricity impact IT's energy consumption, let's consider how we use servers in corporate data centers. In most distributed server architectures, we tend to use large numbers of individual servers, each of which is usually dedicated to a single purpose. There are servers for databases, servers for applications, servers for Web content, and so on.

The more servers you put in the data center, the higher your utility bills climb--both for running the servers and for cooling the data center itself. Also, as the number of servers grows, so does the floor space and facilities devoted to the data center. This also adds to the total cost of data center infrastructure ownership.

Generally speaking, servers run at fairly low overall utilization--often less than 20 percent. Unfortunately, electrons have to flow through a server's circuitry even if the amount of information processing occurring is actually relatively low. So servers keep generating heat anyway. That's why server vendors have been taking a variety of steps to build power management into their products. The idea behind power management is that a server's firmware or operating systems can power them down or put them into some sort of low-power mode when they haven't got a lot to do. This reduces their direct power consumption, as well as the heat they generate--which, in turn, reduces overall cooling requirements but not floor space or other resource requirements.

Power management has its limitations, however. Firmware doesn't have the intelligence to know what is supposed to be running on the box, so it can't determine whether low utilization is really due to a lack of work--or whether it's due to a temporary network problem or load balancing issue. In fact, without some sort of automated workload management, there will always be a substantial inefficiency in the way each individual server's resources--and therefore its direct and cooling-related power requirements--are allocated to the actual computing needs of the business.

The Mainframe's Unmatched Energy Efficiency

Mainframes present IT organizations with a very different profile when it comes to power consumption. In marked contrast to the locomotive-sized systems of the past, the modern mainframe is a model of energy-efficiency and occupies no more floor space than comparable enterprise servers. There are significant "electron economies-of-scale" to be gained by running a small number of high-capacity processors at high utilization in a single mainframe, rather than a large number of distributed servers at much lower utilization, because the CPUs and all of the energy-hungry components surrounding them don't have to be replicated in multiple physical machines.

In fact, a single mainframe contains far less circuitry overall than combined sum of the circuitry in the number of distributed servers usually required to deliver the same amount of business processing. This means that for a given volume of business processing, the mainframe's overall power consumption for both computing and cooling can be as much as 90 percent less than the equivalent set of distributed servers.

In addition, unlike distributed servers, mainframes are able to drive high utilization by:

Running lots of disparate work on the same operating system image
Virtualizing those operating systems so that they can share access to the physical processor resources

This high utilization means that IT can get more computing done within the total fixed energy budget of the data center. Of course, the distributed server market is currently attempting to respond to the utilization problem by offering virtualization technology. The idea is that if you have a lot of server images that have low overall utilization relative to the size of the machine they are currently running on, you can now combine them and run multiple virtual instances of your server operating system on the same box, or deploy an even larger box with even more images on it.

If that sounds like LPAR or VM on a mainframe to you, then you're absolutely right. The leading hardware vendors are beginning to offer fairly primitive LPAR-like partitioning on their enterprise-class servers, but most of the focus in the distributed space is on software virtualization. Software based solutions like VMware's ESX Server and Citrix Systems' XenServer are attempting to accomplish for Windows and Linux what VM did for mainframe systems a generation ago. These solutions are improving, but aren't nearly as mature as zVM mainframe software virtualization.

A major weakness is their relative lack of sophistication and integration between the hypervisor and the hardware stack and between the hypervisor and the guest operating system. Their capacity planning and guest provisioning tools aren't as mature and workload management is almost non-existent. This means that it is quite difficult to ensure that server images are dispatched according to the business importance of the work running within them. IT organizations typically rely on an amalgamation of third-party tools to achieve virtualization capabilities that only partially resemble what's already available on the mainframe. So even though distributed systems are improving steadily, they still can't deliver the same cost and energy efficiencies we currently take for granted with mainframe systems.

How should this affect IT decision-making? For one thing, it should lead IT executives to reconsider their long-term application development strategies. With energy costs rising and corporate "green" initiatives gaining momentum, there are sound business reasons in favor of expanding mainframe investment and using the added capacity to support both growth in existing mainframe applications and to "right-sizing" distributed server deployments onto zLinux and zVM.

Just as important, IT executives have to think strategically for the long term. Energy prices are likely to get even higher in the future. In some locations, there is even a threat of limited power availability--especially in summer months when rolling brownouts are becoming the norm. This makes the power problem an issue of more than a few dollars here and a feather in the cap of corporate responsibility there. Cost savings can run into six figures. And reduced power consumption could mean the difference between service availability and service failure.

You can investigate this issue further yourself by measuring your own systems' power consumption today. Crunch the numbers, and consider the real cost of continuing to build more distributed systems infrastructure indiscriminately. You'll see for yourself that the efficiency of the mainframe makes a compelling case for expanding its role in the data center--instead of reducing it.

Chris Craddock is a 30-year industry veteran with 20 years of commercial product development experience. He is currently senior vice president and distinguished engineer in the Office of the CTO at mainframe software maker CA.