next up previous contents
Next: Timer resolution: TICK1 Up: Low-Level Benchmarks Previous: Performance Range: The

Single-Processor Benchmarks

The single-processor low-level benchmarks provided by PARKBENCH , aim to measure performance parameters that characterise the basic architecture of the computer, and the compiler software through which it is used. For this reason, such benchmarks have also been called appropriately basic architectural benchmarks. Following the methodology of Euroben [11], the aim is that these hardware/compiler parameters will be used in performance formulae that predict the timing and performance of the more complex kernels (see Chapter 4) and compact applications (see Chapter 5). They are therefore a set of synthetic benchmarks contrived to measure theoretical parameters that describe the severity of some overhead or potential bottleneck, or the properties of some item of hardware. Thus RINF1 characterises the basic properties of the arithmetic pipelines by measuring the parameter (\rnhalf) (see subsection 3.2.3), and POLY1 and POLY2 characterise the severity of the memory bottleneck by measuring the parameters (\rfhalf) (see subsection 3.2.4).

The fundamental measurement in any benchmarking is the measurement of elapsed wall-clock time. Because the computer clocks on each processor of a multi-processor parallel computer are not synchronised, all benchmark time measurements must be made with a single clock on one processor of the system. The benchmarks TICK1 and TICK2 have, respectively, been designed to measure the resolution and to check the absolute value of this clock. These benchmarks should be run with satisfactory results before any further benchmark measurements are made.
Tue Nov 14 15:43:14 PST 1995