Sunday, May 10, 2015

Trace Viewer

The Empire Trace Viewer presents a simple UI view that is a facade covering a massive amount of data. As the commercial product runs, and the trace feature is enabled, tracing data is saved in a file that facilitates debugging and performance statistics.  A typical session will generate hundreds of millions of lines of output.
The user story of interest concerns how to process the trace file so the fraction of contents the user wishes to view can be presented with no perceptible delay between dragging the file into the view and seeing the filtered results.  The other, similar user story is, how can the trace be presented with no perceptible delay after one of the filter options is changed.

I handled this problem by loading what is essentially function-keyed hash tables within call ID-keyed hash tables within session-keyed hash tables, which at the leaf end contains a list of file offset addresses.  With that, all lines in the trace file corresponding to which session the user recorded, which call made into the system, and which functional ID's are of interest can be accessed and presented in near immediate time.

No comments:

Post a Comment