shinyGantt 4 StarPU

Collection of python and R scripts aiming at doing post-mortem analysis of an execution of a StarPU program.

dag2pydict takes a dag.dot file and can find all predecessors of a given task.

tasksRec.py converts a tasks.rec to a csv that can be easily read with R. This kind of trace is lighter compared to paje generated trace but shows only tasks events.

paje2csv.sh converts a paje trace to a csv. Richer than tasks.rec generated trace since it contains runtime state (notably sleeping, overhead and mpi communication thread status). pj_dump must be available. (see PajeNG)

gantt.R generates the gantt. (The prepare_trace function should be adapted to your needs and the input. By default, it is adapted for a tasks.rec generated trace.)

shinyGantt allows to display the gantt in a browser. It is possible to zoom using the small gantt. The gantt global variable is used to choose the gantt to display. The DisplayState column is used by default to choose the color of the tasks. When clicking near the start of a task, infos are displayed in the text box.

Typical usage

• Convert tasks.rec to readable csv.

$> python tasksRec.py tasks.rec tasks.csv

• In R, to display the Gantt, assuming all your files are in the same place (scripts and tasks.csv):

> #test
> library(ggplot2)
> library(shiny)
> setwd("path/to/files")
> #Load the gantt
> tmp=read_paje_trace("tasks.csv")
> t=prepare_trace(tmp)
> #Fill the gantt global variable (used by shinyGantt)
> gantt=t
> #Launch the shiny app.
> runApp("shinyGantt")

• If you want to generate a gantt from a paje trace, do the following :

$> ./paje2csv.sh input.paje tasks

Be sure that the prepare_trace function is modified. By default, it is designed to take its input from a tasks.rec-generated csv file.