Application Technologies

We spent a great deal of time in Chapter 11 discussing the three application models (network computing, collaborative computing, and client/server computing) that are major investment areas for Version 4. That investment is expected to continue well beyond Version 4. There is no indication that these application models will be replaced with others in the near future, but then, most of the industry did not predict the rapid acceptance of the network computing model. There will probably be some new application model in the next 5 to 10 years, but it is hard to predict what that might be. The AS/400’s capability to incorporate new models will no doubt be used again. For every possible application to run on any system would be wonderful, but that is not reality. Many AS/400 customers want to run applications that are written for another system, or more precisely, for another operating system. The integrated application servers I described in Chapter 11 allow AS/400 customers to do just that. Specifically, these servers can run applications written for OS/2, AIX, and Windows NT. If other operating systems become important for AS/400 customers, IBM can easily add new integrated servers.

The underlying technology for new application development will be OO. The OO technologies have already demonstrated the potential to greatly increase development productivity, and new applications will be developed with these technologies. Traditional procedural programs will continue to be enhanced until they are redesigned or totally replaced. This process will take many years.

The biggest problem with OO is the skill level it requires. As a result, future application development and customization will take place on many levels, with a range of skills required. For example, only a relatively small group of professionals who create operating systems and application development tools will use the native OO languages such as C++ (otherwise known as the double-edged razor blade). Solution providers and ISVs will likely use languages such as Java, and customizable frameworks and reusable components, such as those from the San Francisco Project. In-house programmers for most businesses will likely use visual component interconnection, while casual users will use human-centered interfaces and self-learning tools.

Before 2001, we will know how well the Java initiative, “write once, run anywhere,” has worked out. With every major platform having implemented a Java virtual machine, the possibility of writing a single Java program that runs on all platforms exists. How universal this will be is a matter of debate. OO technology, in general, provides us with a new paradigm for application development. Only time will tell how quickly this OO adoption actually happens. The direction, however, seems clear to most in the industry.

Total System Performance

Here and in Chapter 11, I have outlined many of the future possibilities for the AS/400, including plans to greatly enhance system performance to meet the needs of tomorrow’s business applications. It should be clear that we are going to create some very high-performance AS/400 systems in the future. But what about today? How does the AS/400e series stack up against its competitors right now?

I was recently looking at a list of performance benchmark numbers for several computer systems and thinking about how we measure performance. We often try to come up with a single number, a number that will tell a prospective buyer everything there is to know about that computer. Of course, we all know that it is ridiculous to base a decision on just one number, but most of us do it anyway.

A convenient number is the megahertz (MHz) rating of the processor. As I have said before, measuring the MHz of a processor is like measuring the RPM of an automobile engine. It tells how fast the engine is spinning, but it says nothing about how much work is being performed. Many processors today are spinning very fast but accomplishing little work. Benchmark programs have been devised to give us information about how much work is really being done.

Benchmark Programs

Many, many benchmark programs are used today to measure performance. With so many benchmarks to choose from, picking a valid one for your business can be a daunting task. A few independent organizations, however, have created benchmarks that many computer vendors support. Probably the two most significant ones are the Standard Performance Evaluation Corporation (SPEC) and the Transaction Processing Performance Council (TPC).

A group of computer vendors formed the SPEC organization in 1988 to create a benchmark suite for workstations and servers in the Unix marketplace. The SPEC benchmark suite is a collection of programs written in C and Fortran. Some of the benchmarks are integer-oriented, from which the SPECint performance is computed, and some are floating-point-oriented, from which the SPECfp performance is computed. Performance is computed by running the programs one at a time and measuring the elapsed times. The geometric mean (multiply n numbers together and take the nth root) of all these elapsed times is calculated to get a final result.

The first set of benchmarks was SPEC89, named after the year in which it was introduced. The original set contained 10 programs (4 integer and 6 floating-point). SPEC92 increased the number of programs to 20, and even more were added for SPEC95, the latest version. Work is now underway on SPEC98.

Because the programs are very small and are run one at a time, an entire program usually fits into the on-chip cache memory of the computer system being measured. SPEC95 did add a few larger programs; but then, cache memories are getting bigger. As a result, SPEC can measure raw processor performance; but because it does not stress the memory or I/O subsystems of the computer, you cannot use it to estimate overall system performance. As a result, SPEC’s primary use is to estimate performance for a single-user Unix workstation. As expected, processors with high MHz ratings, such Digital’s Alpha processors, do well on SPEC benchmarks.

The TPC benchmarks are designed to measure total system performance rather than just processor performance. According to its mission statement, the TPC is a non-profit corporation founded to define transaction processing and database benchmarks and to disseminate objective, verifiable TPC performance data to the industry. There are about 43 members of this organization, including all the major computer vendors.