The Mainframe Can Help Customers Go Green

The System z platform has been answering the call of energy efficiency for decades

The System z platform has been answering the call of energy efficiency for decades
Illustration by Gina and Matt

Building a More Efficient Datacenter

When energy was less expensive, it made sense for datacenters to incrementally add distributed servers as computing needs increased. Each time a company needed a new application, the solution was adding a server to handle that application. The temperature was raised bit by bit without the datacenter administrators realizing they would soon be in hot water.

Soon, all of the servers generated such heat that they necessitated serious investments in cooling equipment, but all this cooling equipment jacked up the datacenter's power needs. "They could fix the cooling, but they'd run out of power," says Anderson. "And if they fixed the power, then they ran out of UPS (Uninterruptible Power Supply) capacity." Adding batteries to protect equipment power generates additional power loads. Adding more servers as computing needs grew only made the problem worse.

"Datacenters that faced incremental growth and added small servers bit by bit did not anticipate the cumulative costs of hundreds or even thousands of servers," says Anderson. "But there's a better to way to build a datacenter, and that's by installing a large centralized server that can take the place of hundreds of smaller less energy-efficient servers."

Anderson explains that building an efficient datacenter begins by prioritizing integration and simplification (see Figure 1). Generally, datacenters with less network equipment and fewer servers are more efficient in energy, cooling and space than their crowded counterparts. A large centralized server like the System z server can take the place of hundreds of small distributed servers. This is especially popular with Linux* applications and servers.

Not only are older distributed server farms power-hungry, according to Anderson, but also most of their power usage goes to the processor and memory, which can be idle most of the time. Anderson likens this to leaving the TV on when nobody is using it. By contrast, mainframes use a major portion of the power for I/O and the processor remains busy doing useful work. The efficiencies in packaging, such as highly efficient power supplies, and the economies of scale inherent in a large centralized server enable mainframes to achieve greater energy efficiencies than a network of distributed servers. By combining many servers into a single server, mainframes also eliminate many of the pieces of power-consuming networking equipment from the datacenter's raised floor. Read the EPA report on this topic.

As the IT industry has evolved, CMOS technology improved. Each generation of CMOS technology normally requires less voltage and can perform better than its predecessors. Recent current leakage in chips, a product of scaling to the atomic levels, caused chip designs to use much more power for performance gains than in previous generations. All state-of-the-art microprocessor designs now must carefully balance cycle time, (performance) with power requirements and cooling capabilities. The System z* server utilizes seamless technology that's been optimized for power performance. The z9 microprocessor using CMOS 10K constitutes a vast improvement over old, power-hungry distributed servers (see Figure 2).

IBM mainframes offer asset protection - a significant advantage in the fast-changing IT environment. A datacenter that relies on distributed servers must throw out its old servers if it wishes to upgrade to newer, more energy-efficient equipment.

A mainframe owner who purchased a z990 mainframe with 130 nm CMOS technology and processors capable of 450 MIPS in 2003 and wishes to upgrade to the latest generation System z9 computer faces a better prospect. The customer may be eligible for credits for MIPS and specialty engines in the old machine. The customer gets a new mainframe with dense 90 nm CMOS technology and engines capable of 600 MIPS that can be scaled up to almost 18,000 MIPS with better watt-per-MIPS efficiency.

"Customers report consistently that hotspots or out-of-power conditions in datacenters are caused by distributed system proliferation, not mainframes." - David F. Anderson, PE, IBM green consultant

Aaron Dalton is a writer who specializes in business and technology topics. Aaron can be reached at

comments powered by Disqus



2018 Solutions Edition

A Comprehensive Online Buyer's Guide to Solutions, Services and Education.


Your Input Needed: IBM Systems Media Reader Survey

IBM Systems Magazine Subscribe Box Read Now Link Subscribe Now Link iPad App Google Play Store
Mainframe News Sign Up Today! Past News Letters