Linux activity monitor
Graphical visualization of the following data:
CPU utilization and per core clock speed
GPU utilization and clock speed (so far only Nvidia). Should work with multiple GPUs (not tested)
Memory and Swap utilization
Network utilization (Wlan and Ethernet). Wlan link bandwidth is constantly updated.
You can also see an overview of running processes.
I wanted to have a graphical visualization of the load of most physical devices. The system monitor in ubuntu does a great job, but is not displaying HDD and SDD load as well as no GPU load. This tools brings all the information together to one place, similar to task manager in windows.
1. Installation and run
Dependencies (Python 3):
pip install pyqtgraph pyqt5
clone the project and cd into src/. Then
If you want to monitor your Nvidia GPU, nvidia-smi has to be installed.
Probably you want some shortcut. To do so first find the path of your python installation:
and then the command would look like
If you want an icon I recommend installing alacarte for this. inside alacarte just click 'New Item', then Name: 'sysmon' and for command the command you found above.
2. Data origin
Most data comes from the /proc directory.
In more detail:
- CPU: /proc/cpuinfo and /proc/stat
- Memory: /proc/meminfo
- Disks: /proc/diskstats
- Network: /proc/net/dev and iwconfig(Wlan)
- GPU: nviida-smi (if Nvidia GPU)
- Processes: 'ps -aux'
I this choice of data source makes the program usable across multiple linux distributions. (Only tested on Ubuntu 18.04 and 20.04)
3. To do
Include support for Intel and AMD GPUs. I don't have any of these GPUs, so unfortunately I can't test it. Maybe I can get my hands on any of these GPUs in the future.
Add more information about Wlan connection. (Signal strength etc.)
Maybe add public and private IP address somewhere
Add more GPU information (Process overview)
Add GUI menu to modify parameters such as update interval, capture period etc.
Add iowait for each CPU core
Add more detailed Memory information (buffer and cache)
CPU clock: Sometimes clock for threads on same physical core differ --> I think this is in reality not possible. Maybe calculate average per physical core. Also group threads belonging to the same physical core together and "bond" visually.
Add dark mode
Make a C/C++ port (probably lower hardware requirements, even though CPU load is low.)
Remove Swap entry if no swap is used
Remove 'ps -aux' references
Add kill button and sort