Competitive Advantage: Mainframe

The objective of this article is to provide a significant, unbiased base of information useful to organizations and to stimulate in-depth review of their IT costs from a new perspective.

IT has never stood still; the industry rapidly changes. Three years ago, we predicted the downfall of mainframes and wrote about distributed server markets in that context. A good hard look at the real numbers showed a different picture. When organizations perform this same in-depth review of their own real numbers, it causes many managers to sit up and take notice.

Competitive Advantage through Lower Costs

The biggest news in IT these days is the placement of new workloads related to the Internet on mainframes and not distributed servers due to the lower total cost per transaction. It’s generally less expensive, by a factor of 10, to place new workloads related to the Internet on mainframes than on distributed systems. Shocking, isn’t it? On an application-by-application and case-by-case basis, the costs are less because mainframes implement shared workloads with utilization rates typically in the 85 percent range while distributed servers in the data center implement single applications, yielding low utilization on a perserver basis.

Enterprises face enormous competitive risk if they persist in building out increasingly expensive data centers by continuously adding to their distributed server farms while their direct competitors move new workloads related to the Internet onto mainframes. Competitive advantage is strategic, and lowering the cost per transaction is fundamental to automated process implementation. Mainframes continue to be the solution to achieve the lowest cost per transaction.

The IBM System z Total Cost of Ownership (TCO) illustrates the significant cost efficiencies that have accrued as IBM enhanced its mainframes over the last decade. When electricity and floor space costs are compared, the mainframe has shown costs that are less by a factor of 100 in an application-by-application analysis. The shared workload capability of mainframes, combined with the need for less labor to fully manage mainframe computing environments, delivers lower overall costs. The result is that mainframes provide a more affordable, network-savvy system that’s better attuned to modern Application Program Interfaces (APIs) and middleware. IBM recently moved much of its own internal workload off distributed servers and onto mainframes, and is continuing to do so solely based on an analysis of the electricity and floor space savings being realized.

The different IT platform choices—mainframe vs. distributed computing servers—need to be compared in a holistic manner. It’s common, but not wise, to simply say, “A server costs $35,000 and a mainframe costs $4 million.” The cost of a typical data center today is $60 million and the cost of a mainframe that supports similar workloads is $6 million. Cost analyses should look at labor, security, hardware and software, network, infrastructure, Service Level Agreements (SLAs), business costs of downtime, and costs of lost business. Industry analysts generally agree that the cost of downtime is about $1 million per minute for a large enterprise.

When platform comparisons are performed in a holistic manner, it no longer makes sense for the large enterprise to measure TCO strictly on a one-applicationper-server basis. TCO must be measured for 10 to 250 applications on one mainframe vs. 10 to 20 blades or a grid of 50 distributed systems. Comparisons performed in this more realistic manner demonstrate that the TCO for mainframe systems is 10 times better than Unix, Linux, Windows, and other similar alternatives. TCO is seven times less expensive when the Linux workload is virtualized on the mainframe. In mixed workload environments, mainframe systems are clearly economically justified.

Key savings also stem from the mainframe’s ability to scale for additional workloads while reducing management and maintenance costs. Mainframes can cost-effectively scale to handle large amounts of additional workload without the need to proportionately increase IT staff to manage the increased processing due to their design and highly reliable, built-in automation systems. By centralizing as many operations as possible on one platform at a single geographical location, rather than managing servers across multiple branches or locations, maintenance expenses can be significantly reduced. By being able to reduce distributed server staff, these valuable human resources can be reassigned to higher-level tasks that help achieve business goals. Labor costs typically represent 40 percent of the data center cost and electricity represents an additional 35 percent. Reducing these costs has a significant impact on a company’s per transaction costs. The ability to have large quantities of different functioning workloads on a mainframe system drives down the labor costs to a fraction of those required for distributed systems.

Mainframe hardware is commonly perceived as more expensive than distributed servers, with upfront costs running about two to three times more. But a closer look at depreciation schedules shows otherwise. A mainframe lasts a decade or more compared with three to five years for a distributed server. Companies allocate those mainframe costs back to departments and others using a chargeback system that’s deceptive in many ways, causing the real costs to be largely hidden from department managers. However, in reality, mainframe depreciation schedules continue for two or three times as long as distributed servers. Organizations often forget this calculation when comparing mainframe and distributed costs.

The TCO costs and the savings that can be realized by moving workloads to IBM’s z9 Integrated Information Processor (zIIP) and System z Application Assist Processor (zAAP) specialty engines on mainframes also are compelling. For Java and selected DB2 workloads, these specialty engines are designed to dramatically lower both hardware and software costs. IBM’s new pricing model for System z New Application License Charges (zNALC) encourages new applications, including Java applications running under WebSphere Application Server (WAS), by providing discounts of 80 to 90 percent on the Monthly License Charges (MLC) of z/OS. Here’s how it works:

zIIP and zAAP specialty engines are priced much lower than the general purpose z processors. A 580 MIPS specialty engine processor is typically $125,000 for an EC class machine and $95,000 for a BC class machine at U.S. list price for a one-time charge, about 9 percent of the price of a general processor. There’s no charge for any IBM software running on the zIIP or zAAP. Typically, upgrades to the next generation of the zIIP and zAAP processors are free.

So, if 50 percent of the workloads are redirected to these specialty engines, savings of 50 percent on the software cost and more than 45 percent on the hardware cost can be achieved.

The zIIP specialty engine is used to redirect selected DB2 workloads. In database server scenarios, up to 40 percent of the queries received via Distributed Relational Database Architecture (DRDA) Remote Access Protocol (RAP) can be redirected. In data warehouse scenarios, up to 80 percent of parallel queries can be redirected. Some of the index maintenance utilities can be redirected to zIIP, too. In an SAP or data warehouse scenario, a 40 percent redirect ratio is typical. zAAP is used to redirect Java workloads. In WebSphere scenarios, up to 85 percent of the workload can be redirected to zAAP. Some customers even report redirecting 97 percent of their workload to zAAP, dramatically lowering software costs.

IDC research proves the TCO of mainframes is much lower than for distributed systems. The IBM System z mainframe offers unmatched reliability, scalability, and security, and helps customers improve the productivity of Java/WebSphere/DB2 applications development. A recent IDC study proves that the TCO of the mainframe is only 56 percent of the cost of distributed systems, considering the costs of hardware, software, services, networking, IT staff, and user downtime.

More cost savings come from IBM’s pricing curves and trade-in policies, lower costs for storage, disaster recovery and energy costs, staffing efficiency, and high utilization. A Java/WebSphere ROI application analysis of the costs of acquisition gives a comparison between System z and distributed systems in this context. The distributed cost is 10 times more than the mainframe cost because the shared workload environment is more efficient than the low utilization on distributed servers.

Shared workload is the most significant aspect of reducing costs. Enterprises should take a hard look at moving new and many current distributed system workloads to the mainframe to reduce costs. The TCO costs—lower electricity and labor costs with higher hardware utilization—and the savings that can be realized by moving workloads to the zIIP and zAAP specialty engines on mainframes are compelling. You should take an independent look at the relative costs of application implementation, as the competitive threat posed by lowering the cost of automated processes directly impacts your bottom line. Staying competitive or gaining a competitive advantage is all about lowering the cost per transaction, and the computing platform used makes a significant difference.