Allow compiled monitors to be auto-configured by training from a sample trace

Issue #28 open
Sean Kauffman repo owner created an issue

The idea is to allow a trace to be passed along with the compiler and then run the spec on the events to determine how large the various caches need to be. The output from the run would be silent and it might even be possible to guess a window size if the nfer algorithm was modified to track statistics on how large of a window was used.

This doesn’t feel terribly difficult, actually. Most of the pieces are there. This just requires putting a few things together.

Comments (2)

  1. Sean Kauffman reporter

    This is partially resolved, in that setting --log-level=2 when analyzing a trace will output suggested memory settings for the compiled monitor when it’s done. The part about analyzing the window size isn’t done, but shouldn’t be hard to add. The functionality is a bit naive, also, and can lead to some bloated memory settings, but maybe that’s okay.

  2. Sean Kauffman reporter
    • changed status to open

    As mentioned in the comment, this is partly resolved. Just need to have it suggest a window size and perhaps a flush threshold.

  3. Log in to comment