2021-2022 SHARCNET Courses

Scheduling software on a compute cluster is how one runs programs/jobs on that compute cluster. When jobs are submitted to the scheduler, the scheduler directs where and when those jobs run within the compute cluster. The order jobs are run is based on each job's priority calculated using that system's fair share policy. Finally, each job that runs has its requested resources tied to their account so it is able to use such when it runs.

This course discusses Compute Canada's compute cluster scheduling software (SLURM) as well as aspects of how jobs are scheduled, using interactive jobs to do (human) interactive (live) work, and how to estimate job resource requirements.

Live Session Dates: Oct. 12, 14, 18, & 21

NOTE: If enrolling after the live session dates, recordings of such can be viewed online.

This introductory course will focus on introducing Machine Learning (ML) in order to run ML code on Compute Canada's clusters. There will be some hands-on exercises and some discussion on how to submit ML jobs to the Graham compute cluster, ML frameworks such a PyTorch, Tensorflow, and Scikit-Learn, and a brief introduction to a ML linear regression model.

Live Session Dates: Nov. 8 & 11.

NOTE: If enrolling after the live session dates, recordings of such can be viewed online.

Writing a concurrent or parallel program that can run on multiple cores or on multiple nodes/computers in a cluster is not the same as writing a sequential program for a single CPU core. Typically sequential programs are only concerned with some aspects of how information is stored in RAM and the step-by-step (sequential) correctness of its code. A concurrent program is concerned with the non-sequential reasoning of how code executes and how it can see (or not) values in RAM from other streams of execution, how it orders reads and writes on those streams of execution, and the time it takes for data in to be transferred across busses (e.g., CPU to GPU RAM, CPU to machine RAM, etc.).

This course will provide a practical overview of important concepts and details involved with concurrent programming in order to further understanding of the benefits, costs, and issues involved with exploiting and writing parallel computing programs.

Live Session Dates: Nov. 22 & 25

NOTE: If enrolling after the live session dates, recordings of such can be viewed online.

Since the introduction of the milestone of Fortran 90 standard, Fortran has become very different from its ancestor with many new modern language features. Inherited from old FORTRAN standards and enhanced by the newer ones, Fortran is still the language of choice for scientific computing for its intrinsic support for real and complex floating point operations. Its array handling, enhanced by index slicing, makes array processing similar or identical to its counterparts, such as MATLAB, hence superior to any other programming languages. The unified interface of intrinsic functions for all data types makes translating mathematical expressions to computer programmes a lot cleaner and easier. Further, since the standard 2008, the introduction of co-arrays has made writing parallel code that can run on multiple core computers and clusters trivial, one needs not to know the traditional message passing interface (MPI) library, nor threading. They are built into the syntax of the language itself.

This course is a companion to the C++ course, where you will find the features missing from Fortran that are more essential to general purpose computing than number crunching. We will show some of the practice of interlanguage programming, through which the shortcomings of Fortran could be complemented by interfacing with C/C++.

Live session dates and time: Monday and Thursday, Jan. 10, 13, 17, 20, 24, 27 at 2pm-3pm EST.

NOTE: If enrolling after the live session dates, recordings of such can be viewed online.

CMake is a cross-platform, open-source software tool used to automate the building, testing, packaging, and installation of software. It is independent of the programming language and tools used although it is most commonly used to with C, C++, and most recently Fortran code. CMake is particularly nice since it determines for a number of programming languages the dependencies between various files, e.g., header and source files, in order to properly build the software/documentation and to minimize the amount of work required to build such.

This course will introduce CMake as well how to write, run, and test CMake scripts. Especially if you are manually building your programs by hand or are experiencing difficulties using shell scripts, it is worth exploring how to use tools such as CMake to make this much easier and much less error-prone to do.

Live Session Dates: Feb. 14, 17, 22 & 24.

Singularity is a container solution similar to Docker for secure use on multi-user high performance and advanced computing (HPC and AC) systems. Singularity has evolved significantly since it was first introduced in 2015 supporting features such as overlay files, image encryption, exposing GPGPU devices to the container software, etc. Features aside,  using containers is valuable for many: the installed software remains exactly the same and can be used a number of clusters and personal computers running Linux by everyone in a research team.

This course will cover commonly used aspects of Singularity, e.g., building an image from DockerHub, how to build a custom image of your own, how to use an overlay file to have read-write access to an image, and how to build and make use of Conda software installations inside an Singularity image.

Live Session Dates: Mar. 7 & 10

NOTE: If enrolling after the live session dates, recordings of such can be viewed online.