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 202 the use of those resources, and is easily controlled in a flexible manner by its user. In particular, a microanalytic simulation systemâthe computing system as seen by the social scientist or policy analystâis useful to the extent that it addresses and provides effective methods of performing key functions. Among the important key functions are â¢ managing the data file(s) containing the population microdata and associated information about them, including codebook documentation; â¢ storing the microdata files and other large files in an efficient and cost-effective manner; â¢ preparing initial populations of microdata from external data sources; â¢ binding internal model variables to external microdata file attributes; â¢ providing a mechanism to define the computer programs that are executable versions of the operating characteristics of the underlying microanalytic model; â¢ providing a mechanism to define entire computer-based model programs by integrating modules that represent individual operating characteristics; â¢ ensuring logical consistency among individual modules or operating characteristics within the model; and â¢ providing a full-function, effective user interface that gives the user flexible control over the functionality of the system. While good computer-based solutions do not yet exist for supporting all of these functions, current developments are addressing a number of these areas and promise considerably more effective computing environments than are now used for microanalytic simulation activities. Advances in hardware and software technology can provide better solutions than have been available in the past. Below, several aspects are addressed of what we believe will be included in future computing environments available for microanalytic simulation activities. The discussion focuses on desktop computing environments because of our belief that they are becoming adequate and appropriate for such activity. Larger Model Execution on the Desktop The ability to run large and complex models is important. In order for microanalytic socioeconomic policy models to be useful in their current contexts, simulation population sizes must be large. The behavioral detail reflected in household economic behavior is sufficiently rich that small samples cannot portray it. In addition, policy proposals generally have a regional, state, or provincial impact that can be important in judging their relative acceptability. Large populations are required to derive acceptable results for smaller areas. On the other hand, for testing and experimentation purposes and for some simulation experiments, large populations are not required. Efficient simulation
FUTURE COMPUTING ENVIRONMENTS FOR MICROSIMULATION MODELING 203 systems support data organizations that provide the flexibility of choosing appropriate sample sizes so that small samples can be used when appropriate.56 It should be apparent that the desktop hardware platforms likely to be available by 1995 will be capable of supporting microanalytic simulation models of considerable size. With single processors expected to have speeds of 40â100 MIPS, it is clear that the speed of executing microanalytic simulation runs is no longer a significant issue. To the extent that parallel processor architecture becomes available on the desktop in such a way that tasks can be easily divided to run in parallel, the increase in speed will be multiples of what is anticipated for a single processor. The large simulation population size required, combined with the repetitive use of such models in social science research, requires that the simulation population be organized in as efficient a form as possible. In the past, microanalytic simulation systems have met this requirement by compressing the data in different ways, including variable field width storage, bit-packed bytes and words, tagged fields, and zero field suppression. Two costs are involved: the cost of storing the data on an external medium such as a disk or tape and the cost of reading and then converting the data into internal form for the simulation program. In the past, storage devices available for attachment to microcomputer systems did not have the capacity or the flexibility of their mainframe counterparts. The processing of socioeconomic data for microanalysis or microsimulation depends on processing large data files, often in parallel. The large data storage alternatives of magnetic tape and large on-line disk storage provided by mainframe computer systems have given them a relative advantage historically in meeting this need. The desktop environments of 1995 are likely to retain the need for some type of compaction, but the amount of memory available should suffice for simulation runs involving even large populations. With primary (immediate-access) memories of 32â64 MB expected to be common, compacted files of the size currently processed by TRIM2 may be able to reside within primary memory during the course of a simulation. With respect to secondary storage devices, current developments in microcomputer-based systems indicate that secondary desktop storage in 1995 will be at least as good as secondary storage on current medium-sized mainframes if accounted for on a per-job basis. Low-cost on-line disk storage of 40â300 56 SPSD/M's capabilities in this regard are quite effective. The simulation population is ordered in such a manner that the first sequential 5, 10, and 25 percent of the population records are stratified samples of the entire population and therefore constitute efficient samples. Furthermore, within those intervals, the population units are randomly ordered, so that the first N households always form an unbiased sample of the entire simulation population. The SPSM system allows for use of the first P percent of the population. In addition, it provides a break mechanism allowing a user to interrupt and terminate the simulation at an arbitrary point. SPSM then weights the partial aggregate tables to obtain correct results for the universe.