Computation steering is a scientific and technical
area which provides methods of understanding a state of
running high-performance computing (HPC) programs and
performing interactive control over such programs. This work
is an analysis of ideas and current status of computation
steering. The main idea of computation steering is that
computation and visualization are performed simultaneously.
By the way, modern HPC systems approached the disbalance:
they have much more computing power than storage power.
Thus computations cannot output all meaningful data easily.
To solve this, steering technologies are used – with idea to run
in non-interactive mode, extracting program’s in-memory data
and saving it for ordinal post-hoc analysis. This lead to new
demands on algorithms of correct data reduction, in purpose of
correctness of various following data analysis. Another idea of
modern steering is a configuration of parts of visualization
pipeline in text files separate from simulation codes. This allows
to relatively easy setup data transformations and connections
between them and their placement on various locations,
including computational nodes. Modern steering systems rely
on model: describe simulation data’s life using API and then
configure in text files what to do with that data in purpose of
visualization during simulation.

https://ieeexplore.ieee.org/document/9117685