Parallel Processing Performance

Allied with design and preparation, catalyst testing is the exploratory screening of candidate catalysts. This phase does not yield either kinetics or process variables but merely ranks performance. Bench reactors used should be as simple and rapid as possible, for many samples are usually tested. For ease in operation and interpretation, model compound reactions are helpful. Thus, for example, cumene dealkylation is a model for catalytic cracking, and thiophene hydrogenolysts for hydrodesulfurization. Care must be taken to ensure that the model system does indeed parallel process performance. 

As CPU performance increases, the gap between CPU and disk and memory speeds will continue to widen. As limits of technology are approached, other techniques will be needed to gain performance advantages more functional units, multiple processors, and so on. These approaches are discussed in the sections on minisupercomputers and parallel processing. 

When the reaction is performed at relatively low temperatures that prevent strong thermal decomposition of the alkali metal carbonate, the formation of C02 will be related only to the reaction and will indicate the stoichiometry of the process. Fig 8 presents mass loss isotherms of Nb02F - M2CO3 mixtures (in which M - Li, Na, K, Rb, Cs) that were subjected to thermal treatment in air at 850°C [84, 85]. It is important to mention that parallel experiments performed without the addition of Nb02F, resulted in alkali metal carbonate mass losses that were in the same order of magnitude as the measurement errors at temperatures below 850°C. 

For another investigation, amide formation was used as a model reaction to demonstrate the performance of parallel processing in micro-channel devices [23]. The target of such processing is combinatorial synthesis, the provision of multiple substances within one run. 

The issue of parallel versus sequential synthesis using multimode or monomode cavities, respectively, deserves special comment. While the parallel set-up allows for a considerably higher throughput achievable in the relatively short timeframe of a microwave-enhanced chemical reaction, the individual control over each reaction vessel in terms of reaction temperature/pressure is limited. In the parallel mode, all reaction vessels are exposed to the same irradiation conditions. In order to ensure similar temperatures in each vessel, the same volume of the identical solvent should be used in each reaction vessel because of the dielectric properties involved [86]. As an alternative to parallel processing, the automated sequential synthesis of libraries can be a viable strategy if small focused libraries (20-200 compounds) need to be prepared. Irradiating each individual reaction vessel separately gives better control over the reaction parameters and allows for the rapid optimization of reaction conditions. For the preparation of relatively small libraries, where delicate chemistries are to be performed, the sequential format may be preferable. This is discussed in more detail in Chapter 5. 

According to the DARPA Neural Network Study [18] A neural network is a system composed of many simple processing elements operating in parallel whose function is determined by network structure, connection strengths, and the processing performed at computing elements or nodes. ... 

These levels of mental processing continue in parallel and often support each other. Each mental process can result in different types of performance with different impacts on process equipment. One of the main successes of the S-R-K model was its prediction of previously unrecognized mental process performance levels and the problems associated with using the wrong level. The original S-R-K model has been modified and extended by several researchers, but it remains a useful concept for relating human performance to specific (potential) root causes. 

Processing steps after harvesting may be carried out simultaneously in the same production area provided that adequate precautions are taken to prevent cross-contamination. For killed vaccines and toxoids, such parallel processing should only be performed after inactivation of the culture or after detoxification. 

Whereas the criteria (i) enantioselectivity, (ii) amount of product obtained per amount of catalyst used, and (iv) substrate specificity are of a quantitative nature, the point (iii) availability (though not cost) of a catalyst is only a semi-quantita-tive criterion, and (v) comparison of a method with alternative strategies is even redundant, as the process of comparison of options and their resulting evaluation and ranking are merely different parts of the decision-making process, and are not undertaken in parallel with performance measurements along dimensions of merit. 

Hsu, T.T., On the Performance and Cost Effectiveness of Some Multiprocessor Systems, 1977 International Conference on Parallel Processing, August 1977. 

Shared-memory parallel processing was certainly more successful for QC in earlier applications and continues to play a significant role in high performance computational chemistry. A coarse-grained parallel implementation scheme for the direct SCF method by Liithi et al. allowed for a near-asymptotic speed-up involving a very low parallelization overhead without compromising the vector performance of vector-parallel architectures. 

---