CAMNAS.jl

Context-Aware Modified Nodal Analysis Solver Plugin for the Dynamic Power System Simulator DPsim.

CAMNAS is a simple Julia implementation of a Modified Nodal Analysis (MNA) solver designed for integration as a solver plugin in DPsim. Its key features include:

• Automatic detection of NVidia GPU accelerators
• Dynamic switching of accelerators between time steps
• Basic rule-based automation for accelerator selection during runtime (under development)
• Asynchronous system monitoring/decision-making for accelerator selection (under development)

Building

git clone https://github.com/RWTH-ACS/CAMNAS.jl.git
cd CAMNAS.jl
julia --project=$(pwd) --eval="using Pkg;Pkg.instantiate()"
make

Dependencies

Build Depedencies:

• julia>=1.11
• gcc
• make
• ...

For an overview of the required Julia package depdendencies, see the [deps] section in Project.toml.

Usage

<dpsim-scenario> -U "Plugin" -P "camnasjl"

To allow CAMNAS and DPsim to dynamically link the necessary libraries, ensure that your LD_LIBRARY_PATH includes both the CAMNAS base directory and CAMNASCompiled/lib/.

For more information on the interaction of DPsim and the Plugin, see DPsim's MNASolverPlugin.cpp.

Environment Variables

Certain features and behavior patterns of the solver can be managed through environment variables during the initialization of the plugin. For details, refer to the following:

Variable	Values [default]	[TYPE] Description
JL_MNA_DISABLE_AWARENESS	Boolean [false]	[CONTROL] Toggle accelerater detection/awareness. When disabled, calculations will always fallback to CPU.
JL_MNA_RUNTIME_SWITCH	Boolean [false]	[CONTROL] Allow migration between accelerators during runtime.
	Allow Accelerator Types
JL_MNA_ALLOW_CPU	Boolean [true]	[CONTROL] Allow accelerator selection to use CPU.
JL_MNA_ALLOW_GPU	Boolean [true]	[CONTROL] Allow accelerator selection to use GPU.
	Accelerator Forcing
JL_MNA_FORCE_CPU	Boolean [false]	[CONTROL] Forces the use of CPU. Ignores other strategies!
JL_MNA_FORCE_GPU	Boolean [false]	[CONTROL] Forces the use of GPU. Ignores other strategies!
	Accelerator Selection Strategies
JL_MNA_ALLOW_STRATEGIES	Boolean [true]	[CONTROL] Allows the plugin to use strategies for opinionated accelerator selection.
JL_MNA_LOWER_POWER_STRATEGY	Boolean [false]	[CONTROL] Selects accelerator with lowest estimated power consumption (based on TDP values).
JL_MNA_HIGEHEST_FLOP_STRATEGY	Boolean [false]	[CONTROL] Selects accelerator with highest estimated FLOPs value.
JL_MNA_SPECIFIC_ACCELERATOR_STRATEGY	Boolean [false]	[CONTROL] Selects accelerator specified by JL_MNA_SPECIFIC_ACCELERATOR.
JL_MNA_SPECIFIC_ACCELERATOR	Boolean [nothing]	[CONTROL] Name of the accelerator to select. (Requires JL_MNA_SPECIFIC_ACCELERATOR_STRATEGY)
	Utils
JL_MNA_PRINT_ACCELERATOR	Boolean [false]	[DEBUG] Print currently used accelerator independent from debug statements.

Warning

To modify the plugin's behavior interactively during runtime (e.g., dynamically switching accelerators between timesteps), users can edit the automatically generated system.env file, or directly modify CAMNAS internal variables using CAMNAS.update_varDict!(key,value). However, this feature is still experimental, therefore not recommended, and requires to set the optional ENV variable JL_MNA_RUNTIME_SWITCH before loading CAMNAS.

Development / Testing

Developing or testing a shared library or plugin can be tideous, especially due to long compilations times. However, the purely Julia script test/test_interface.jl comes to your help!

This script allows testing all externally accessible functions directly within Julia.

To ensure compatibility with DPsim, three test case inputs of different sizes are provided. The specific input size can be adjusted using the inputType variable at the beginning of the script. All inputs are designed to satisfy the equation x = A \ b, where A is a sparse matrix in CSR format and b is a vector.

Contribution

We require all commits to be signed-off by the contributer. By doing this, you state that you agree to the terms of the "Developer Certificate of Origin (DCO)" (https://developercertificate.org), for the submitted contribution.

Author

• Felix Wege fwege@eonerc.rwth-aachen.de

Institute for Automation of Complex Power Systems (ACS) EON Energy Research Center (EONERC) RWTH University Aachen, Germany