Quickstart
This quickstart shows how to enable the JUmPER IPython extension in a notebook, start monitoring, and inspect basic performance data for a few cells.
All examples below assume that jumper_extension has been installed as described in the Installation section.
Enable the extension
In a Jupyter notebook or IPython shell, load the extension once per session:
%load_ext jumper_extension
This registers the %perfmonitor_* magic commands and wires them into the underlying monitoring service.
Minimal monitoring workflow
1. Start monitoring
Begin collecting performance data for subsequent cells:
%perfmonitor_start [interval]
intervalis optional and specifies how often metrics are sampled in seconds.- If omitted, the default interval of
1second is used.
2. Run your code
Execute the cells you want to profile as usual. While monitoring is active, JUmPER records CPU, memory, GPU, and I/O metrics over time.
3. View a performance report
Show an aggregate report for the current session:
%perfmonitor_perfreport
%perfmonitor_perfreport --cell 2:5 --level user
- Without arguments, the report covers all cells executed so far.
--cell RANGErestricts the analysis to specific cells (for example5,2:8,:5, or3:).--level LEVELselects the monitoring scope:process,user,system, orslurm(if available).
The report prints aggregated metrics such as CPU utilization, memory usage, GPU utilization, and GPU memory across the selected range.
4. Plot performance data
Open an interactive plot with widgets for exploring metrics over time:
%perfmonitor_plot
The plot lets you:
- Zoom into interesting regions of the timeline.
- Filter by cell ranges.
- Switch between monitoring levels.
You can also use a direct, non-widget mode and export results:
%perfmonitor_plot --metrics cpu_summary,memory
%perfmonitor_plot --metrics cpu_summary --level user --cell 2:5
%perfmonitor_plot --metrics cpu_summary,memory --save-jpeg performance_analysis.jpg
%perfmonitor_plot --metrics cpu_summary --level user --pickle analysis_data.pkl
--metricsaccepts a comma‑separated list of metric keys such ascpu_summary,memory,io_read,io_write,gpu_util_summary,gpu_band_summary, andgpu_mem_summary.--save-jpegwrites the current view to an image file.--pickleexports the plot data for later interactive analysis.
5. Inspect cell execution history
Review all executed cells with their timestamps and durations:
%cell_history
This opens an interactive table that lets you correlate individual cells with collected performance metrics.
6. Stop monitoring
When you are done collecting data, stop the monitor:
%perfmonitor_stop
One‑command fast setup
For a fully configured environment with interactive plotting and automatic reports, use the fast setup command:
%perfmonitor_fast_setup
This command:
- Enables
ipympl‑based interactive plots in the current notebook. - Starts the performance monitor with a
1.0second interval. - Enables automatic performance reports after each cell at the
processlevel.
For detailed options and additional commands, see the Public API section.