Parallel programs performance measures

We describe in this section various terms and concepts pertaining to parallel computing, including the classification of parallel computers, issues relevant to development of parallel algorithms, and measures for assessing the performance of parallel programs. 

Once a program has been implemented, measurement-based evaluation is useful to refine its implementation by highlighting sources of inefficiency like load imbalance or high communication costs. Performance visualization, based on recorded program behavior, is a particularly powerful method for detecting unexpected behavior within a parallel program. 

In this chapter we will consider issues pertaining to parallel performance modeling. We first introduce some network performance characteristics for parallel computers that must be considered when modeling parallel performance. We then present several performance measures for parallel programs, and we discuss how to develop a performance model for a parallel algorithm. Finally, we will discuss how to evaluate performance data and illustrate how reported performance data can be potentially misleading. 

Performance Measures for Parallel Programs 5.2.1 Speedup and Efficiency... 

One of the most widely used performance measures for parallel programs is the speedup, S(p), which is defined as... 

These steps are rarely performed in such rigid lock-step manner but all of the actions described must be accomplished prior to successful completion of executing a real parallel problem on a cluster. The effectiveness achieved in programming a cluster is difficult to measure (although some metrics have been devised to this end). Nonetheless, the ease of parallel programming is strongly influenced by the execution model assumed and the tools available to assist in the process. 

Performance and scaling of the simulator program were measured via parallel efficiency. Parallel efficiency refers to the ratio of the ideal run time to the actual run time. For example, if the perfeet speed-up would have given a run time of 10 seconds and the actual run time was 12 seconds, then the parallel efficiency is 10/12, or 83.3%. Care must be taken while measuring parallel efficiencies so that there are no other jobs running on the cluster that would share the CPU time with the benchmark runs. 

An alternative to quantitative analysis by ICP-MS is semiquantitative analysis, which is generally considered as a rapid multielement survey tool with accuracies in the range 30-50%. Semiquantitative analysis is based on the use of a predefined response table for all the elements and a computer program that can interpret the mass spectrum and correct spectral Interferences. This approach has been successfully applied to different types of samples. The software developed to perform semiquantitative analysis has evolved in parallel with the instrumentation and, today, accuracy values better than 10% have been reported by several authors, even competing with typical ones obtained by quantitative analysis. The development of a semiquantitative procedure for multielemental analysis with ICP-MS requires the evaluation of the molar response curve in the ICP-MS system (variation of sensitivity as a function of the mass of the measured isotope) [17]. Additionally, in the development of a reliable semiquantitative method, some mathematical approaches should be employed in order to estimate the ionisation conditions in the plasma, its use to correct for ionisation degrees and the correction of mass-dependent matrix interferences. 

Experiments were performed on an automated parallel synthesis workstation coupled with a personal computer supplied with software enabling the workstation to be programmed. Samples were prepared in a rack containing a 6x4 array of glass tubes. Catalytic activities were measured by means of gas chromatography. The reported activities are the averages of the results from different experiments. 

The proposed method is based upon the quantitative measurement of the contribution of differently charged nitroxide probes to the spin-lattice relaxation rate (1/T i) of protons in a particular molecule, followed by the calculation of local electrostatic potential using the classical Debye equation (Likhtenshtein et al., 1999 Glaser et al., 2000). In parallel, the theoretical calculation of potential distribution with the use of the MacSpartan Plus 1.0 program has been performed. 

---