You have just received your Compute Canada account. Welcome! Now what do you do? This page is intended to help you find your way through the technical documentation on Compute Canada services and systems.
|If you don't already have a Compute Canada account, see Apply for an account.|
What do you want to do?
- If you are an experienced HPC user and are ready to log onto a cluster, you probably want to know
- If you are new to HPC or would like some training, you can
- If you want to know which software and hardware are available for a specific discipline, a series of discipline guides is in preparation. At this time, you can consult the guides on
- If you have hundreds of gigabytes of data to move across the network, read about the Globus file transfer services.
- Python users can learn how to install modules in a virtual environment and R users how to install packages.
- If you want to experiment with software that doesn’t run well on our traditional HPC systems, please read about Compute Canada Cloud resources.
- If you prefer a video webinar format, please check our Getting Started with the new National Systems mini-webinar series.
What resources are available?
Compute Canada is currently installing several million dollars’ worth of new computers while simultaneously retiring many old computers. During the transition period (2016-2018), a changing mix of old and new computers will be available to you. You can read about the progress of the migration from old to new systems here.
New resources (deployed in 2016 or after)
Arbutus is an extension of the West cloud. Arbutus went into service in September 2016.
Cedar and Graham are general purpose clusters composed of a variety of nodes including large memory nodes and nodes with accelerators. They went into service in the summer of 2017. You can log in to either one using SSH and the same password you use at ccdb.computecanada.ca. A home directory will be automatically created for you the first time you log in.
Niagara is a homogeneous cluster designed for large parallel jobs (>1000 cores). It entered service in March 2018.
Your password to log in to all new national systems are the same one you use to log in to ccdb.computecanada.ca. Your username will be displayed at the top of your home page at ccdb.computecanada.ca once you've logged in there.
Legacy resources (deployed before 2016)
Computing and storage resources which were installed between 2004 and 2015 and scheduled to be decommissioned in the next few years are referred to as legacy resources. The legacy resources are administered by regional organizations, one of ACENET, the Centre for Advanced Computing, Calcul Québec, SciNet, SHARCNET, and WestGrid. To use a legacy resource you must have an account with one of these entities; you can apply for an account through CCDB. Resources deployed during and after 2016 will not require this step, nor will the two clouds.
Most legacy clusters are classified as either capacity clusters or capability clusters. Capacity clusters contain nodes connected to each other by a relatively slow Ethernet network, while the capability clusters have a fast network, usually InfiniBand. Large parallel jobs will run better on capability clusters than capacity clusters, while smaller jobs will run almost anywhere.
- Helios and Guillimin, hosted by Calcul Québec
- Accelerator Research Cluster, hosted by SciNet
All of these have NVidia GPUs. Guillimin also has Intel Xeon Phis.
What resources should I use?
This question is hard to answer because of the range of needs Compute Canada serves, and because of the enormous variety of resources we have available --- especially during the 2016-2018 renewal period. If the descriptions above are insufficient, contact Compute Canada’s technical support or your regional support.
In order to identify the best resource to use, we may ask specific questions, such as:
- What software do you want to use?
- Does the software require a commercial license?
- Can the software be used non-interactively? That is, can it be controlled from a file prepared prior to its execution rather than through the graphical interface?
- Can it run on the Linux operating system?
- How much memory, time, computing power, accelerators, storage, network bandwidth and so forth --- are required by a typical job? Rough estimates are fine.
- How frequently will you need to run this type of job?
You may know the answer to these questions or not. If you do not, our technical support team is there to help you find the answers. Then they will be able to direct you to the most appropriate resources for your needs.