Nowadays, many computationally intensive codes, such as those used for physics or biology simulations, are run on GPU-accelerated hardware. On large computing clusters, this hardware is distributed across multiple nodes, which communicate with each other via Message Passing Interface (MPI). The goal of ISAAC is to visualize this data without the need for expensive copy processes between accelerator and host. This allows ISAAC to fully take advantage of the high computation speed of these accelerators.
ISAAC consists of three parts: The server, the in-situ library, and the client.
Simulation code just has to add some calls and settings to the in-situ template library. After that, the server will notice when a simulation is running and give the user some options to observe the computations on the fly via a client. The client is able to show the transferred stream and meta data. It is also possible to send meta data back to the simulation, e.g., to restart it with improved settings. Multiple clients can connect to the same server and observe the streamed visualizations.
Please see in INSTALL.md for installing, building and using ISAAC and its dependencies. If you need to install ISAAC on a server not accessible from the outside, you need to tunnel the connections of the clients. A more detailed documentation about using ISAAC can be found here (currently outdated).
A reference HTML5 client is provided here, but needs to be adapted to specific
simulations.
Inside the client directory, the HTML client visualisation.html can be found.
It lets you connect to your visualizations directly from your web browser and allows to change settings and steer the visualization. For visualizations of PIConGPU with ISAAC enabled, it will also display meta data like the cell count or current time step.
As an alternative, a Python client for Jupyter Notebooks is also in development and can be found here.
ISAAC is licensed under the LGPLv3.
