Description
Students should form a group of up to 4 students that will work on choosing a topic (some suggestions appear below). Most commonly, the project should involve developing a parallel program, either for a shared-memory platform, a distributed memory platform, or a massively parallel SPMD platform such as a GPU. Feel free to propose your own project; we are available to discuss any ideas you have.
To ensure that the topic is well-matched to the course material and that it is of appropriate difficulty, you can contact the instructor and TAs, and discuss the feedback of your submitted abstract.
1 Deliverables
The three deliverables for the project are listed below.
Project abstract (10%): Submit a brief description of the project you will undertake on
MyCourses by the prescribed deadline.
Presentation (40%): Presentation in class the during the last two weeks of the course. You are expected to be present for all days of those lectures to present when called and also to
help in completing and evaluating your classmates’ presentations.
Code and Writeup (50%): The code and a write-up summarizing the results of your project are due by the midnight of the last day of the classes.
2 Project Abstract
Submit a brief project description. The document should include a summary of your project and justification for why the project is appropriate for four people to complete in the remaining half of the semester. Your project description should include:
Problem statement: What problem will you be solving?
Methodology: What approach will you take to solving this problem? What type of platform will your program require, and what programming framework will you use?
The following two parts are not required in the abstract submission, but it is good for you to discuss them as you finalize the proposal.
Testing and/or evaluation: How will you test your program to demonstrate it is accurate and to gauge its performance?
Workload breakdown: Who will be responsible for the different tasks/aspects of the project?
Be as specific as possible in answering all of these questions.
3 Presentation
In such a project, a serial motion estimation algorithm would be used as the implementation base.
A first step of optimization would be the usage of common loop transformations, for parallel processing capabilities to be revealed. The next step would be the parallelization of these code segments using either multicore programming tools (e.g OpenMP, MPI etc) or GPGPU tools (e.g CUDA, OpenCL etc). More than one parallel implementations could be explored and an extensive comparison between different approaches could be presented.
5.3 Architectural Considerations in Multiprocessors on Chip and Networks on Chip (NoC) Over the past 5 years, Networks on a Chip have grown in importance. Many applications could benefit from both the Software and Hardware resources that a NoC offers. Different architectural considerations could be explored using efficient open-source simulation tools (e.g NoC simulators).
In this type of a project, the execution of a sample parallel application (e.g Gaussian elimination) on a NoC will be inspected. The performance impact of varying different architectural parameters should be evaluated. For example, the impact of routing variability on performance could be explored. Furthermore, the impact of different data decomposition schemes or interprocess communication patterns (e.g point-to-point, collective) on the memory transactions overhead could be inspected.
5.4 NVIDIA Jetson Board Usage
This platform promises to deliver very power efficient parallel computing in a small package.
In this project, the different programming tools (e.g libraries, APIs etc) should be extensively explored. Representative parallel implementations should be used as simple testing scenarios.
5.5 Mobile Computing – High performance Architectures
Parallel and distributed computing advances have led to complex hardware systems that enable high performance applications on commercial electronics (i.e. smartphones, tables, etc.). For instance, Apple’s Metal API provides efficient access to the GPU, maximizing the graphics and compute potential of your iOS applications.
Several hardware acceleration mechanisms have been developed and integrated to recent smartphone motherboards. For example, recent Android APIs enable hardware acceleration at the application, activity, or window level.
In this project, existing mobile computing APIs should be explored. Simple iOS- or Android-based applications should be developed that exploit such mechanisms. Performance increase using mobile GPUs for representative functionalities (i.e. scaling, rotating images) should be reported.
Different mobile OS and architectures could be tested. Comparison between newest and older systems (i.e. current versus previous Android OS editions) could be presented.
5.6 Quantum Computing Exporation
Quantum computing principles provide extensive parallelism via quantum entanglement and state superposition, and there is currently lots of research being undertaken in making quantum computing a reality. Projects in quantum computing can take form of exploring the basic principles, tracking the published progress, trying out quantum computing resources (e.g., offering available on Amazon Web Services), or some algorithmic developments (e.g., Quantum Fourier Transform, cryptographic protocols, quantum key distribution, …).
