Why is an exponential distribution used for the call times?
• This distribution is commonly used for modeling call times. Although there are better models (e.g., log-normal), this one is extremely easy to generate and gives an appropriate spread of precisions in the input numbers. • Only about 16 bits (4 digits) are needed for the input numbers; why use 64 bits (15 digits)? • Partly to make sure the benchmark doesn’t underestimate the amount of data to be read from disk, and partly to allow the use of a ‘generic’ random number distribution data file. • Why does the package require the use of decimal arithmetic (if available)? • Decimal arithmetic can be used to assure the exactness of the results (see the Decimal FAQ). Also, using a decimal data type reflects actual applications of this type (the data are often held in decimal form, and using decimals improves maintainability compared to manual scaling of integers). Note that the format of the input file (binary or packed decimal) may be chosen to suit the representation used by the decimal ari
