Today we have the following:

• Telemetry to capture time taken to start jupyter
  • We draw conclusions that this is how long it takes for user to run a cell
  • This is not correct, as we now start jupyter server in the background. Assume starting jupyter is 10 seconds, 30 seconds later user opens an nb and runs a cell. Cell would run in <1s, but telemetry reports 10s.
  • We have identified this as we're now using ExecuteCellPerceivedCold/Warm telemetry for nb
• Using ExecuteCellPerceivedCold/Warm to draw the conclusion on how long it takes to run the first cell is still inaccurate
  • Assume majority of the users were to run a large complex cell that takes 20s.
  • This telemetry would report >20s, when in fact jupyter could start in 3s!

Suggestion (more accurate)

1. Create new telemetry to capture PerceivedJupyterStartup

• In code we check time taken in InteractiveBase.submitCode from start of method to after this._notebook.executeObservable. We do this for first time code is executed.
• I believe this accurately represents time taken for nb to be ready to execute user cells:
• How long it takes for jupyter to start before it is ready to run user code (in notebook)

2. Create new telemetry to capture StartExecuteCellPerceivedCold

• We could then make it more accurate to determine time taken for Jupyter to start running a user cell.
• Previous suggestion + time till Jupyter reports it is busy (when its busy we know it is executing user code).

• I think this is what we're after today, i.e. how long does it take our code to start running user code from the time they hit run

@jmew @rchiodo @IanMatthewHuff @DavidKutu /cc