Postgraduate Course: Advanced Message-passing Programming (INFR11169)
|School||School of Informatics
||College||College of Science and Engineering
|Credit level (Normal year taken)||SCQF Level 11 (Postgraduate)
||Availability||Available to all students
|Summary||Modern parallel supercomputers are predominantly programmed using the Message-Passing Interface (MPI) library. Learning the basic MPI syntax and writing medium-scale programs is relatively straightforward, but many issues only arise when tackling large-scale problems on thousands of processes. Typical issues include deadlock, poor scalability or inefficient file IO.
There are two basic ways to address these issues. The fundamental MPI calls can be used in more sophisticated ways, which requires an in-depth understanding of the finer details of the MPI standard and its implementation in real libraries. Alternatively, different approaches can be used which employ advanced MPI functionality or which exploit the shared-memory nature of modern multicore compute nodes. Before any correctness or performance issues can be addressed they must first be diagnosed, so knowledge of parallel profiling and debugging is essential.
The course will cover the following topics:
- Scalability challenges for parallel programs
- MPI Internals
- Parallel profiling and debugging
- Message-passing optimisations
- Advanced communicator management
- Advanced use of MPI collectives
- Hybrid message-passing and shared-memory programming
- Unstructured meshes
- Advanced load-balancing techniques
- Parallel file systems and parallel IO
Entry Requirements (not applicable to Visiting Students)
|| Students MUST have passed:
Message-passing Programming (INFR11163)
It is RECOMMENDED that students have passed
Threaded Programming (INFR11178)
||Other requirements|| Relevant OpenMP prerequisite material will be made available for self-study, but will not be taught on this course.
Ability to program in C or Fortran
Information for Visiting Students
|Pre-requisites||Message-Passing Programming / MPI - Mandatory
Threaded Programming - Recommended (relevant material will be made available for self-study, but will not be taught on this course)
Ability to program in C or Fortran.
|High Demand Course?
Course Delivery Information
|Academic year 2018/19, Available to all students (SV1)
|Learning and Teaching activities (Further Info)
Lecture Hours 20,
Supervised Practical/Workshop/Studio Hours 10,
Summative Assessment Hours 2,
Programme Level Learning and Teaching Hours 2,
Directed Learning and Independent Learning Hours
|Assessment (Further Info)
|Additional Information (Assessment)
||Via practical classes.
||Hours & Minutes
|Main Exam Diet S2 (April/May)||Advanced Message-passing Programming||2:00|
On completion of this course, the student will be able to:
- Describe the various factors that limit performance in large-scale message-passing applications
- Apply knowledge of the design and implementation of the MPI library to optimise the performance of parallel programs
- Use advanced MPI functionality to exploit the architectural features of modern HPC systems
- Implement efficient parallel approaches for general unstructured problems
- Compare the pros and cons of different approaches to parallel IO
|"Using Advanced MPI: Modern Features of the Message-Passing Interface", William Gropp, Torsten Hoefler, Rajeev Thakur and Ewing Lusk.|
Any further material provided via Learn.
|Graduate Attributes and Skills
||Ability to address complex problems using methodical approaches.
Application of learning.
Ability to understand and apply technical documentation.
||There are limited spaces on this course. Students not on the MSc in High Performance Computing or MSc High Performance Computing with Data Science should contact the course secretary to confirm availability and confirm that they have the required prerequisites before being enrolled on the course.
The course is available to PhD students for class-only study. PhD students requiring a form of assessment (e.g. SUPA/School of Physics and Astronomy CDT students) must contact the course secretary to confirm method of enrolment.
|Additional Class Delivery Information
||2 lectures and 1 practical per week (Weeks 1-10).
|Keywords||Advanced MPI,MPI,Parallelism,EPCC,HPC,Message-passing,Programming,High Performance Computing
|Course organiser||Dr David Henty
Tel: (0131 6)50 5960
|Course secretary||Mr Ben Morse
Tel: (0131 6)51 3398