I have seen many research papers about parallelizing an existing serial solution. This trend is quite common in the field of computer graphics. Is parallel processing of a scalar solution really considered as a novelty?
Efficient parallelization of existing scalar algorithm is a nontrivial job, so i would say yes, parallel processing of a scalar solution can be considered as a novelty, but if and only if you can point out an interesting and/or not obvious problem that has been solved while doing the project. At least you should consider several various parallelization approaches.
I think we can have novelty at various levels:
1) Developing a new parallel algorithm (not just simple parallelisation of an existing serial algorithm)
2) Developing a parallelisation system, or a general approach for parallelising a class of problems
3) Providing for novel parallel programming models / synchronisation semantics
Yes,
just take solution of linear system Ax=b
Good parallelization is quite a complicated task.
Similarly if a serial solution is medu with a different mathematical approach offering extension to a higher dimension or new way for parallelization etc.
Parallelization of scalar problems is indeed a trend in the research community, especially over the course of the last decade. To address your question, I have a few topics I'd like to discuss:
1 - Certain algorithms/problems that were not used in real-world applications because of their complexity/high computational load have been shown to be real-time when deployed in a parallel architecture. These make excellent candidates for parallel programming research efforts.
2 - There are many problems whose parallelization is quite trivial. These are great examples for researchers looking to dive into this topic. However, most of them have already been studied, so there is no novelty there. However, problems whose characteristics makes them non-trivial for parallelization are still a challenge nowadays, and it is a undoubtedly a novelty.
3 - Parallel programming is highly dependent on the target architecture, as there is no "best" code that suits all architectures. Many works are flawed in this aspect. A good comparison of parallel approaches over different architectures should be optimized towards each architecture (even devices from the same family), and these differences reported.
In conclusion, if you identify a way to parallelize a non-trivial problem, and if you compare different platforms, and optimize your application towards each one, reporting the results and what you did to achieve the performance, I'd say you definitely have novelty.
- Badges
- Science topic
- Similar topics
- Papers