In contrast to the familiar situation where a user runs e.g. Word on his or her personal laptop, a supercomputer may serve hundreds or thousands of users (as an example, CSC has currently over two thousand users) who connect to the system through the internet. Also, where most laptop users utilize a graphical user interface (i.e start applications by double-clicking with a mouse and generally use a mouse extensively), the most common way to use supercomputers is via command line. So, instead of using a mouse, one works by typing commands with a keyboard. Even though the command line interface might at first feel arcane to someone who has only used graphical desktop applications, the command line is a very powerful and flexible way to use the computer. After all, supercomputers are to normal computers what a formula 1 car is to a common stationwagon, great power may come with more difficult usage. However, where very few people have a chance to even try to learn how to drive formula 1, any body can learn the basics of using supercomputers via command line in one day.
Example of a terminal with two spereate instances
Left: Batch file ; Rigth: Batch job submission
Supercomputers are not made more difficult to use on purpose, and supercomputing centers are constantly looking for ways to make supercomputers more accessible. Nowadays, some supercomputing centers provide also web browser based access, which allows users to utilize supercomputers with more graphical interfaces via web browser (Firefox, Chrome, Safari, Edge, ...) running on their local laptop or desktop computer.
As a supercomputer may have thousands of users, there are not necessarily enough resources (e.g. CPUs, GPUs and memory) for every one to run their jobs at the same time. Thus, users typically cannot run their heavy computations directly whenever they want (this would be called interactive use) but have to submit their jobs through a batch queueing system. The batch system then takes care of scheduling distribution of the resources in a fair way. If the supercomputer has enough free resources, the job can start immediately, otherwise it may take hours or sometimes even days for the job to start. Batch system ensures also that single users cannot monopolize the system, there are typically limits on how much resources a single user can utilize at the time. Batch system makes the utilization of supercomputer more efficient. With only interactive use there would be load mainly during normal working hours, but with the help of a batch system jobs can be run on supercomputer 24/7 throught the whole year.
When using the batch system, the user describes to the system (in most cases in a special file called a batch script) which executable is to be run, how many nodes, cores, possibly GPUs, how much memory is to be used, and finally, how much time is to be reserved for the job. The batch queueing system then uses a complicated algorithm to decide when and in which subset of the nodes the job will be run. Typically, the more nodes and other resources and/or time a job requires, the longer it must wait in queue to wait for its turn to be executed.
Link to additional material: Using command line in CSC supercomputers
Most supercomputing centers all over the world have a similar policy for accessing supercomputers. Supercomputing centers typically have public national funding, and people performing academic research within the country hosting the supercomputers can use them for free. As an example, CSC provides free access to researchers (and sometimes also for students) with an affiliation to a Finnish higher education institution (Universities, Universities of Applied Sciences), or a state research institute, for research where results are made publicly available. Supercomputers are typically available also for commercial usage, however, commercial usage is normally separately invoiced.
Many supercomputing centers participate in transnational access programs where also users outside the hosting country are provided resources in the supercomputer based on special applications (and possibly with scientific criteria for approving applications).
The forthcoming LUMI supercomputer is funded by a consortium of ten European countries (Finland, Belgium, the Czech Republic, Denmark, Estonia, Iceland, Norway, Poland, Sweden, and Switzerland) and European Union. Half of the resources of LUMI is allocated to consortium members, other half is available also for researchers and industries outside the consortium via a special application process.