Introduction

Since you are reading this article, it is assumed that you have an intuitive definition of the term "Mainframe". At one time, the term accurately described the physical characteristics-a complex collection of highly specialized electronics, all bundled together in a set of monolithic boxes and used as the main focus for information technology. The role of the mainframe has not changed. It provides the central focus for an enterprise’s information technology. On the other hand, the market place has spawned a very different hardware/software solution for providing mainframe characteristics, but packaged in a surprisingly non-traditional manner. The technology that was once considered the embodiment of the personal computer has now emerged as a powerful, scalable, high-performance force in the mainframe market.

There was a time when IBM was the dominant force in computing with a huge percentage of the overall computer market share, consisting of large "mainframes". There were other mainframe vendors, but they could not encroach on the market share for a number of reasons.

One of these reasons rested in the overall involvement of many third parties in the "value chain"-the various bits and pieces that add value to the basic hardware and thus provide a complete solution for the customer. The IBM value chain included many large software companies that provided turn key software solutions, along with a bevy of consultants to apply the hardware / software to provide a complete solution for a large enterprise. The IBM hardware was "proprietary", but there was a large "commodity" market consisting of several suppliers of peripherals, supplies and all of the other components that are needed to make an IBM environment successful. As more vendors participated in the ecosystem, the prices dropped due to more widespread computerization, the growing variety of competitors and the high volume that allows for cheaper manufacturing. This ecosystem continued to grow; with success begetting success until the IBM dominance appeared perpetual.

Although the general IBM ecosystem flourished through the commoditization of various supporting components, the main central electronic complex (CEC) was strictly proprietary, with IBM dictating the prices of the new machines. The first wave of disruptive technology for mainframes emerged in the form of UNIX. UNIX promised to be "Open" in that many hardware vendors would manufacture hardware that collectively would support a single operating system. The promise was that, as various vendors achieved higher performance, the IT manager could incorporate their new, cheaper hardware into the complex but continue to support a homogeneous environment. A new value chain emerged, with a new ecosystem to provide all of the necessary pieces that are needed to establish a complete solution.

The cost continued to drop and the raw performance continued to climb. These two facts are not independent. There is a causal relationship between lower cost and higher CPU performance. With the lowering of cost, the adoption of computerization becomes more widespread. In the UNIX case, computerization spread to the departmental level. It was no longer a corporate-wide decision to buy another computer system. Again, the commoditization created a "snowball" effect, with momentum growing as unit sales climbed and prices dropped. The higher sales volume created more competition and provided more R&D dollars to invest in more efficient and higher performing CPU architectures. In retrospect, UNIX was not quite ready for the "big time" when it was first adopted for corporate (as opposed to academic) deployment. But the hardware, OS and ecosystem matured until UNIX became the solution of choice for new enterprise applications. By 1999, it was accepted that UNIX could be used to implement a large, data-intensive computing environment:

Here is a quote from a white paper published in 1999 by Database Associates International [1]:

This paper has demonstrated that a S/390 server running OS/390 and DB2 UDB for OS/390 is a powerful combination for supporting a large data warehousing system. The only real, viable alternatives to S/390 are large UNIX, AS/400 and niche hardware parallel servers. Microsoft Windows NT has yet to demonstrate the scalability and functionality required for such a warehousing system.

The New Wave

In the fall of 2000, Unisys announced the availability of the industry’s first 32-way Intel processor architecture that could take full advantage of the new Windows 2000 Datacenter operating system. This marked the beginning of a new disruptive mainframe technology that would, within three years, cause the industry to refer to UNIX as a "proprietary" operating system. Since the announcement, a number of large vendors, including HP, IBM, NEC and Fujitsu, have entered the market place. Today, four of the top ten leaders in the TPC-C benchmark (non-clustered systems) achieved their transaction rates using Intel CPUs and the Microsoft operating system, with SQL Server as its database environment.

Probably each of you have experienced, over the last one or two years, the effects of CPU commoditization. If you have replaced your personal computer within the last year, you know that you can get from two to five times the performance at a lower cost than you could buy two years ago. At the time that this article is being written, you can order a 2.2 GHz system with 256 Mbytes of memory, 17" monitor and 30 GB of disk for under $600.

This is caused by commoditization. As more and more CPUs are purchased, the revenue can be plowed back into R&D to fit more transistors into a smaller real estate. The higher volumes drive down manufacturing costs per unit. This causes memory prices to drop and Cpu cycles to climb. Unisys, along with other vendors, has taken advantage of this shift in the market to engineer high performance systems using commodity components. To achieve a single OS image, running on a 32 CPU Symmetric Processing (SMP) platform, required a lot of ingenuity and hard work.

Edison once described genius as 1% inspiration and 99% perspiration. It may look simple enough on a white board to draw a 32-way CPU Windows-based system, but there has been (and continues to be) a number of less visible technical achievements that contribute to this final outcome.

Manufacturing and delivering a reliable, enterprise-quality system using 32 Intel processors involves a very close partnership between the hardware vendors and Intel. Intel provides building blocks of up to 4 CPUs on a chip. Combining these with 64 GB of memory, 96 PCI slots and an efficient method to move data from memory to the individual processors is the subject of another paper. It should suffice to say that there are many ways to tackle the problem, with each method requiring its own level of cost/risk analysis. The key here is that hardware vendors do not have to spend their R&D dollars on developing the CPU. They rely on a single commodity supplier that can stoke the performance improvements with the revenue that is achieved through higher and higher sales volume.

In addition to the CPU, consider the commodity operating system and environmental software - Windows with SQL Server. Microsoft worked very closely with Unisys and other vendors to leverage these vendors’ experience in developing operating systems and environmental software for large SMP systems. Critical code was analyzed and rewritten so that processes and threads could execute in parallel on multiple processors. Again, this could be the subject of another paper. A number of tools and techniques were developed to focus on the serialized portions of the code and then to provide high-performance locking and thread management. Below are the published results from the Transaction Processing Council, of the TPC-C benchmarks, as published by the various vendors, for non-clustered systems. The numbers show how the Windows operating system, along with SQL Server, continues to scale up to meet the higher and higher performance of emerging CPU technologies.

But success cannot always be measured by benchmarked results. Therefore, the reader is referred to an article about the deployment of Microsoft’s World Wide Marketing Database [2]. At the time the article was published, the database was "only" 1.8 Terabytes in size. Since then, it "grew some". Certainly, the article from Database Associates International could not anticipate the environment that has emerged over the last 3 years.

Figure 1. tmpC Results, UNIX

Figure 2. tmpC Results, Windows

What’s Different?

One might wonder about the differentiation from one vendor to another in this brave new world where processors and operating systems are no longer developed by the vendor. As mentioned previously, there are significant innovations in the architecture that is constructed around the CPU chips. This includes innovations in cache coherence, memory interconnects and I/O handling. Another key differentiator is established on the basis of overall cost, including the cost of the initial system and the total cost of ownership. The savvy vendor is now focusing its ingenuity and intellectual leadership in new methods to create the whole system at lower cost. At the same time, high availability provides a key differentiator. A modern mainframe must deliver mainframe availability. Finally, simplicity in the area of administration and deployment can reap a significant reduction in the total cost of ownership. Therefore, the new "Modern Mainframe" must provide a complete solution that can be totally available with the ability to grow and support an enterprise that must constantly change to meet its commitments to its customers.

This paper has discussed some significant changes in the mainframe platform over the last decade. The "Windows Mainframe" is just beginning to burgeon into the market place. The concept of commoditization as a force in elevating the performance of CPUs in a mainframe environment was discussed. Although the "Windows Mainframe" has been technically available for three years through Unisys, its recent adoption by other major vendors indicates a validation of the market place and an anticipation for a shift in the pre-conceived ideas regarding enterprise-class computing and "mainframes".

[1] IBM System/390 and DB2 UDB: A high Performance Data Warehouse Server, May 1999 - http://www.databaseassociates.com/pdf/IBMos390%20Version%202.pdf

[2] Scaling Up to the Needs of the Worldwide Marketing Database with Datacenter Server - http://www.microsoft.com/windows2000/datacenter/evaluation/casestudies/unisys.asp