No abstract

Benchmarks that measure time values using a standard system clock often employ a dual loop design. One of the important assumptions of this design is that textually identical loop statements will take the same amount ottime to execute. This assumption was tested on two bare computers with Ada--et programs and has beenL demonstrated to be inaccurate in these specific test cases. J' * Dual Loop Benchmarks: Purpose and AssumptionsBenchmarks are tests designed to measure the capabilities of a computer system. They are used to compare different computer systems and determine the suitability of a computer system for particular tasks. Benchmarks show a wide variation in what they are intended to measure, ( how they are designed, and how they are implemented. Many benchmarks produce outputs that are measurements of the time required to perform some task. A common technique is to write a program that performs some interesting bit of work (e.g., a rendezvous) sandwiched between calls to a system timer.Benchmarks that use time as a unit of measure vary widely in the time they require to run. Some tasks are brief and can take fractional parts of a second to execute. Others measure durations of minutes or hours. The ability of computer clocks to measure this range of times also varies widely. A system clock available to a benchmark designer may be accurate only to a tenth of a second, far too slow to measure an event in the millisecond or microsecond range. More accurate timing devices are often available, but as an option rather than as a standard component. For benchmarks intended for general use, a dual loop benchmark design is often used to permit the benchmark to execute on an unmodified system.Dual loop benchmarks handle the problem of imprecise clocks by extending the duration of the test to a length that the clock can readily measure. The time required for a test is extended by repeating the test numerous times between calls to the system timer. Repetition is usually programmed by inserting the test in a loop, where the number of repetitions may be conveniently changed. The increased time duration of the test series can be measured easily, and the time for the individual test can be determined by computing the average value for the test series. Introducing a loop construct into the test adds time, which must be factored out. This is done with a second loop, a control loop, which contains only the loop construct and not the actual test. The time required for the benchmark is assumed to be the value obtained by subtracting the control loop time from the test loop time. An Ada skeleton for a dual loop benchmark appears as follows:1 'This Ada program fragment requires that the constant SOME-VYALUE and a procedure TEST be added before the program can actually be run, In actual practice, precautions must be taken to ensure optimization by the compiler does not after the essential program structure. For example, because the empty loop contains no executable statements, it might be removed by a compiler. CMU/SEI AVE...

No abstract

Benchmarks are often used to describe the performance of computer systems. This report considers factors that may cause Ada benchmarks to produce inaccurate results. Included are examples from the ongoing benchmarking efforts of the Ada Embedded Systems Testbed (AEST) Project using bare target computers with several Ada compilers.

No abstract