Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
FUTURE COMPUTING ENVIRONMENTS FOR MICROSIMULATION MODELING 197 We believe that while the establishment of a quality-adjusted price index is a worthwhile step in measuring improvements in overall computer industry productivity, the initial series understates the rate of improvement for the purposes of this discussion. Analysis of Industry Information A better estimate would be provided by a specific analysis of price and performance data specifically related to desktop environments for the implementation of systems supporting microanalytic simulation models and modeling. There are a variety of potential sources for such data, including specific manufacturers of components and systems, industry trade groups, industry consulting groups, data collected by trade publications, and private studies. While such data are available in theory, the time required to search for, locate, and analyze them was beyond the scope of the present inquiry. Assessment and Pre dictions Predicting technical progress in the computing industry is tricky. Nevertheless, such progress does occur and is important in understanding for which activities computer-based implementations are likely to be feasible in the future. Of interest to the present authors are desktop markets, including systems that are generally classified in the categories of personal computers, microcomputers, and workstations. These markets are characterized by low barriers to entry,50 high price elasticity of demand, mass markets, large economies of scale in production (especially in software), and very high return from successful innovation. Because of this, we expect a higher than average rate of productivity increase, as innovations create expanding markets and prices drop rapidly with competitive entries and economies of scale. And, although it is not possible to accurately predict the rate of productivity increase, we anticipate that it will be 20â30 percent per year for the next 5â10 years. Productivity increases of 20â30 percent per year imply increases of 74â88 percent over the next 6 years. Let us assume that market options will allow us to take these gains in increased computational power and capacity (instead of lower cost), and let us use as a current system base a 20-MHz Intel 80386 system with 4 MB of primary memory and 60 MB of disk memory. Such a system can be purchased at a discount for about $5,000. Assuming a 90 percent drop in the price-productivity ratio, in 1995 it should be possible to purchase a desktop system with 40 MIPS of power, 40 MB of primary storage, and 600 50 Barriers to entry are higher in some subsystem manufacturing sectors, such as advanced chip technology and high- capacity storage devices. Nevertheless, these sectors are characterized by multiple firms and a high level of competitiveness.
FUTURE COMPUTING ENVIRONMENTS FOR MICROSIMULATION MODELING 198 MB of disk storage for the same $5,000 in real terms. An inflation rate of about 5 percent per year would raise the cost of the system to $6,000â$6,500. Given current industry trends, we believe this is a conservative forecast. It does not take into account changes in hardware or software architecture. It assumes that current engineering improvements can be made at a consistent rate and that no fundamental physical barriers to engineering improvements are reached. It also assumes that all elements of cost are subject to the same rate of technical progress. We conclude that while the 3â100M machine51 will not be here by 1995, it will be close enough for all practical purposes to make the following assumptions with regard to implementing and using microanalytic simulation models: ⢠The processing power available will be sufficient to drive even the most complex graphical user interface and CASE tools for model construction and analysis, as well as to provide more than adequate power for large simulation experiments. ⢠Primary memory will be sufficiently large to be able to store many microanalytic populations, eliminating input and output delays and making random access to micropopulation units feasible. This will help eliminate modeling artifacts, especially in dynamic models, that were required for disk-based micropopulations. ⢠Data compression techniques will still be necessary, since the research appetites of users will grow as capabilities grow, although probably not at the same rate. The amount of substantive and institutional knowledge and the existence of large initial populations for simulation will emerge as more fundamental limitations to model growth as technical capability expands geometrically. ⢠Secondary disk storage will be adequate for juggling a number of large micropopulations as well as large operating characteristic and entity libraries that will be important aspects of the next generation of systems that support microanalytic simulation. In summary, while capacity will probably never be sufficient to respond to all desires of every user, by 1995 desktop systems should be able to support comfortably all of the microsimulation activities performed today while providing much faster response and execution at a fraction of today's prices. Together with the anticipated software developments that we expect and that are mentioned in the next section, the computing environment will be significantly more supportive of and effective for microanalytic simulation modeling activities. The challenge, as discussed in the concluding section, is to adopt 51 The 3â100M machine is characterized by 100 MIPS of processing power, 100 MB of primary memory, and 1 gigabyte of disk storage.
