dacapo benchmarks

[rgrig]

Check a few properties on the dacapo test-suite.

• Express interesting multi-object properties.
• Measure the overhead.

[rgrig]

Some observations after a few runs of an instrumented Tomcat.

Running time goes up 6 times. This is bad. A little profiling shows that the bottlenecks seem to be the functional queue and the use of HashMap. The queue has size <=2, so the current implementation can definitely be optimized a lot. This is good. For the HashMap, I have to look into who exactly uses it too much.

[rgrig]

The top two users for HashMap seem to be (1) the isObservable check and (2) adding to the set of active states. Point (1) is again good news because we can certainly reduce constant factors a lot, by going from hashes to bitmaps. Not sure about (2).

[rgrig]

The overhead is about 3.3% when (1) the checker is inactive, and (2) ~3 interfaces of Tomcat are instrumented.

[rgrig]

I started writing properties for each of eclipse, tomcat, h2, batik, jython, pmd.

[rgrig]

We now have 18 properties in total for DaCapo.