Multi-platform monitoring of the power consumption of hardware and software.
Purpose
As energy is a finite resource and its generation is impacting our climate (whether generating electricity or building devices' batteries), it is crucial to optimize the energy consumption of computing devices.
The first step in optimizing is to measure and estimate the power consumption of hardware and software. In particular, complex devices do not execute and consume power unless tasked by software (OS, applications, etc.), and therefore it is important to isolate the power consumption of individual software.
PowerJoular is a command line software to monitor, in real time, the power consumption of software and hardware components.
It is a multi-platform software capable of measuring or estimating the power consumption of various hardware and software: Intel/AMD x86/64 bits CPUs in servers and computers, Raspberry Pi ARM CPUs, and various other single-board computers (Tinkboard, BeagleBone).
Currently, we support :
PC/Servers using a RAPL supported Intel processor (since Sandy Bridge) or a RAPL supported AMD processor (since Ryzen), and optionally an Nvidia graphic card, on GNU/Linux.
Raspberry Pi devices (multiple models) on GNU/Linux:
Model Zero W (rev 1.1), for 32 bits OS
Model 1 B (rev 2), for 32 bits OS
Model 1 B+ (rev 1.2), for 32 bits OS
Model 2 B (rev 1.1), for 32 bits OS
Model 3 B (rev 1.2), for 32 bits OS
Model 3 B+ (rev 1.3), for 32 bits OS
Model 4 B (rev 1.1, and rev 1.2), for both 32 bits and 64 bits OS
Model 400 (rev 1.0), for 64 bits OS
(soon) TinkerBoard
(soon) BeagleBone X15
PowerJoular is also capable to estimate the power consumption of individual software based on their CPU utilization.
It also offers a systemd service for automatic monitoring, and outputs to the terminal or to CSV files.
Usage
PowerJoular can be used either independently, or its units reused and integrated to other Ada software.
To user PowerJoular independently, just run: powerjoular in the terminal. That's it!
The following options are available:
-h: show the help message,
-p pid: specifiy a particular PID (process ID) to monitor an individual program/process,
-f filename: save monitoring data to the given filename path,
-o filename: save only the last monitoring data to the given filename path (the file is overwritten with only the latest power measures),
-t: print data to the terminal,
-u: update Raspberry Pi (and soon TinkBoard and BeagleBone) power models online (saves to /etc/powerjoular/powerjoular_models.json). Requires root/sudo. Models are downloaded from our public gitlab repository,
-l: use linear regression models (less accurate than the default polynomial models) for Raspberry Pi (and soon TinkBoard and BeagleBone) energy models.
You can mix options, i.e., powerjoular -tp 144 will monitor PID 144 and will print to the terminal.
PowerJoular units can be easily reused as we define functions, procedures and datatypes that are self-contained.
For instance, if wanting to monitor Raspberry Pi CPU power in your 3rd-party Ada application, just integrate the appropriate unit (in this case, Intel_RAPL_sysfs). Same for Raspberry Pi models (dRaspberry_Pi_CPU_Formula):
-- Function to calculate CPU power consumption based on CPU utilization
function Calculate_CPU_Power (CPU_Utilization : Float; Platform_Name : String; PowerModels_Data : Unbounded_String; Algorithm_Name : String) return Float;
Design
Ada's choice to develop and implement PowerJoular was straightforward: a capable language with strong typing, and very low power overhead (ranking among the best in terms of energy efficiency with C, C++ and Rust).
As PowerJoular would be running on low-powerful remote devices, and in many cases continuously, we aimed to limit its energy impact and also limit the possibilities of crashed or bugs.
PowerJoular is designed with 2 main unit categories in mind:
Power measuring or estimating units,
Communicating and distributing measurements units.
For the former, we built individual units for each device category: RAPL CPUs, Raspberry Pi CPUs, NVIDIA GPUs, etc., along with helper units to calculate CPU statistics and processes' CPU usage.
For the latter, we built CSV and terminal unit exports, and we added a systemd service for continuous monitoring.
The architecture design of PowerJoular is further described in this paper.
License
PowerJoular is licensed under the GNU GPL 3 license only (GPL-3.0-only).
Multi-platform monitoring of the power consumption of hardware and software.
Purpose
As energy is a finite resource and its generation is impacting our climate (whether generating electricity or building devices' batteries), it is crucial to optimize the energy consumption of computing devices. The first step in optimizing is to measure and estimate the power consumption of hardware and software. In particular, complex devices do not execute and consume power unless tasked by software (OS, applications, etc.), and therefore it is important to isolate the power consumption of individual software.
PowerJoular is a command line software to monitor, in real time, the power consumption of software and hardware components. It is a multi-platform software capable of measuring or estimating the power consumption of various hardware and software: Intel/AMD x86/64 bits CPUs in servers and computers, Raspberry Pi ARM CPUs, and various other single-board computers (Tinkboard, BeagleBone).
Currently, we support :
PowerJoular is also capable to estimate the power consumption of individual software based on their CPU utilization. It also offers a systemd service for automatic monitoring, and outputs to the terminal or to CSV files.
Usage
PowerJoular can be used either independently, or its units reused and integrated to other Ada software.
To user PowerJoular independently, just run:
powerjoularin the terminal. That's it!The following options are available:
-h: show the help message,-p pid: specifiy a particular PID (process ID) to monitor an individual program/process,-f filename: save monitoring data to the given filename path,-o filename: save only the last monitoring data to the given filename path (the file is overwritten with only the latest power measures),-t: print data to the terminal,-u: update Raspberry Pi (and soon TinkBoard and BeagleBone) power models online (saves to /etc/powerjoular/powerjoular_models.json). Requires root/sudo. Models are downloaded from our public gitlab repository,-l: use linear regression models (less accurate than the default polynomial models) for Raspberry Pi (and soon TinkBoard and BeagleBone) energy models.You can mix options, i.e.,
powerjoular -tp 144will monitor PID 144 and will print to the terminal.PowerJoular units can be easily reused as we define functions, procedures and datatypes that are self-contained. For instance, if wanting to monitor Raspberry Pi CPU power in your 3rd-party Ada application, just integrate the appropriate unit (in this case,
Intel_RAPL_sysfs). Same for Raspberry Pi models (dRaspberry_Pi_CPU_Formula):Design
Ada's choice to develop and implement PowerJoular was straightforward: a capable language with strong typing, and very low power overhead (ranking among the best in terms of energy efficiency with C, C++ and Rust). As PowerJoular would be running on low-powerful remote devices, and in many cases continuously, we aimed to limit its energy impact and also limit the possibilities of crashed or bugs.
PowerJoular is designed with 2 main unit categories in mind:
For the former, we built individual units for each device category: RAPL CPUs, Raspberry Pi CPUs, NVIDIA GPUs, etc., along with helper units to calculate CPU statistics and processes' CPU usage. For the latter, we built CSV and terminal unit exports, and we added a systemd service for continuous monitoring.
The architecture design of PowerJoular is further described in this paper.
License
PowerJoular is licensed under the GNU GPL 3 license only (GPL-3.0-only).