Datamodel

Organize the plethora of quantities into entities.

Classification considerations.

1. parameters for which changes demand restart of the worker
   • geometry file
   • geometry selection
   • flavor of the worker, non-vbo OR vbo
   • propagation kernel
2. parameters of kernel launch, easily tuned
   • threads per block
   • max blocks, determines how the work gets split into kernel launches
3. result timings
   • investigate scaling with number of photons to see what makes sense for comparing different events

Performance Data Model examples

MySQL Performance Schema

Fields:

• instrument name
• time start, time end
• http://dev.mysql.com/doc/refman/5.6/en/events-waits-current-table.html
• http://marcalff.blogspot.tw/2010/01/performance-schema-overview.html

Instrumenting Python Code

• code changes too fast to make line numbers a good may of instrumenting better to name waypoints in the code
• http://stackoverflow.com/questions/4691575/how-to-print-source-code-lines-in-python-logger

__file__,
inspect.getlineno(inspect.currentframe())-1))

Profiling Python

1. lineprofiler-

Projects for instrumentation

• http://www.codeproject.com/Articles/634644/Building-a-code-instrumentation-library-with-Pytho

Graphite : Graphing infrastructure

carbon - a Twisted daemon that listens for time-series data whisper - a simple database library for storing time-series data (similar in design to RRD) graphite webapp - A Django webapp that renders graphs on-demand using Cairo

• http://graphite.readthedocs.org/en/latest/overview.html
• https://codeascraft.com/2011/02/15/measure-anything-measure-everything/

whisper

• https://github.com/graphite-project/whisper/