What is “Niagara at Scale”?
Organized since 2021, “Niagara at Scale” are events where the Niagara supercomputer is reserved for several days for large parallel computations on the order of the size of the cluster.
The fourth installment of “Niagara at Scale” event will be held from October 31st, 2023 at noon EDT until November 3rd, 2023, noon EDT.
Purpose of “Niagara at Scale”
These events enable pre-approved projects that require all or nearly all of the capacity of the Niagara supercomputer at once. Such heroic computations are Niagara’s mandate, as it is the “Large Parallel” cluster within the national systems of the Digital Research Alliance of Canada, and one of the fastest machines of its kind in Canada according to the TOP500 List. But computations of this size — think massively parallel codes running on tens of thousands of cores — are hard or impossible to run within the regular batch scheduler.
How to apply
All Niagara users that have computations which are not trivially parallel, and can only run at such scales, are encouraged to apply to take part in this event.
Announcements of this event went out to Niagara users on September 22, 2023, with an application deadline of October 9, 2023.
Applications should include the intended computation, as well as the number, size (in nodes), and duration of the jobs to be run at scale, as well as the total required /scratch storage space, and are to be emailed to: support AT scinet DOT utoronto DOT ca. Successful proposals will need to show evidence that their codes can run efficiently on at least 10,000 cores on Niagara and include strong and/or weak scaling data and plots, In addition, the codes must be able to checkpoint and restart, especially since jobs will be restricted to a shorter wall time.
Furthermore, if you have a code that does not scale to 10,000 cores yet, but otherwise would qualify, please contact us too, as we may be able set aside some time for scaling and performance analyses in the event as well.
Future events
Niagara at Scales events are held once or twice a year. If you are a Niagara user that has massively parallel jobs or workflows that could take advantage of this opportunity, keep an eye out for future announcements.