Benchmarks

[Baltoli]

In order to evaluate the success of the project, it needs a collection of benchmarks that demonstrate performance benefits:

• Micro / synthetic benchmarks - designed to trigger the particular code path that's being asserted repeatedly and deliberately.
• Macrobenchmarks - should be more representative of "real" code (for example, compiling Clang on an instrumented kernel).

The narrative of the evaluation can usefully include these benchmarks as a motivation and justification of success.

[Baltoli]

I have results from a micro-benchmark where threads contend on an array and do some sorting - it shows a small but noticeable speedup when TESLA instrumentation is removed.

Macro-benchmarks are still blocked on getting stuff from Jon.