Computação Paralela e Distribuída

Esta review é especialmente relevante para quem estuda no Taguspark e precisa de fazer CPD para cumprir o minor, tendo de se deslocar à Alameda.

O projeto em si é relativamente pequeno, as deadlines são razoáveis quando comparadas com outras cadeiras e a matéria lecionada é interessante e relevante. O principal problema surge para os alunos do Taguspark que têm a maioria das cadeiras lá, porque os horários tornam praticamente impossível assistir às aulas de CPD. No meu caso, mesmo com o melhor horário possível, apenas conseguia ir a uma aula prática por semana.

Noutras cadeiras isso costuma...

Projeto bastante pequeno em termos de código que é preciso desenvolver. A avaliação foca-se mais nas qualidade desse código e nas técnicas que usamos para melhorar a performance. Sugiro começar a testar o projeto no deucalion assim que possível porque a fila do deucalion pode demorar várias horas (ou até dias quando está muito cheia) a correr o vosso projeto. O professor da cadeira é competente a ensinar mas a correção do exame é escandalosa, num curso com quase 200 pessoas apenas alguns tiveram mais que 14 no exame... Para estudar são disponibilizados vários exames dos anos anteriores mas...

Projeto giro e facil.

Exame mega lixado por mais que estudes

Nao gostei do professor, parece que ficava chateado com tudo (sabem aqueles videos a tentar mostrar como sao stores do tecnico? Tal e qual).

Dica: ir as aulas praticas q o stor vai dizendo o q melhorar no projeto

Ele tem imensos exames dos anos anteriores

Main classes and themes of the course were interesting, them being OpenMP and OpenMPI. In my year there was a 3 part project that was very light, but I believe it being so light was an exception. It was quite fun to test our projects in Deucalion, but can feel quite a chore. Exams are hard in my opinion, there are a lot of little things and formulas that must be memorised, as professor doesn't give you any options of support In-Exam. Project Discussions were awfull, professor was unreasonably hard/strict on them. Overall interesting course.

Classes → Professor José Monteiro explains the topics clearly, provides a good number of examples, and clearly enjoys teaching the subject. TIP: Taking notes on how computation and communication complexity is calculated across the various examples helps quite a bit (last four classes), the remaining slides explain the remaining concepts clearly.

Exam → The average grade was low — around 11 or 12 if I had to guess, possibly lower. You need to know a lot of small details about OpenMP (especially) and MPI, plus all the other course material, which is a lot of content in my...

The course is not very well organized.

The project was about parallelizing an algorithm, and it had two deliveries. In the first delivery, we had to use OpenMP. We had plenty of time and it wasn't that complicated. For the second delivery we had to use MPI, and not only did we have way less time, teachers also asked us to handle very large inputs. Handling those cases felt like a whole new project.

To test your project, you have to use the RNL cluster, which gets very crowded in the last days before the deadline. The professor's solution to that was basically 'Finish earlier'. -_-

Another...

Project

Historically, this course on parallel algorithms struggles to find good problems to parallelize. A few years back, the project was about parallelizing a Sudoku solver (hint: all the sequential approaches were faster). This year, the only parallelizable part was that the input data was a tree. Other than that, there wasn’t much going on.

Syllabus

The subjects addressed in this course (OpenMP and MPI) are totally outdated. OpenMP has been totally outclassed by things like work-stealing thread pools (see: Rayon or Cilk). MPI is too flexible to really scale well. (For context: our...