Author malte
Recipients jendrik, malte
Date 2017-08-31.23:47:21
If it weren't for the systematic PDB results, I think this would be fine to merge.

I'd like to look a bit more carefully at what goes on in this configuration. The
results for woodworking-opt08-strips in this configuration, where runtime
doubles for all solved tasks, give me pause. Plus 700 seconds just because of
the new hash function? That's a lot. How does this happen when in our
microbenchmarks the new hash function was at worst a few percentage points
slower than the old one? (And of course the planner does other things than
computing hash values, too.)

So two paths to proceed:
1) Remove the changes that make the difference in systematic PDBs, verify that
this configuration is back to old performance, merge, then look at the
systematic PDB stuff in a separate issue.
2) Look at the systematic PDB stuff right now, merge everything in one go.
Date User Action Args
2017-08-31 23:47:21maltesetmessageid: <>
2017-08-31 23:47:21maltesetrecipients: + malte, jendrik
2017-08-31 23:47:21maltelinkissue731 messages
2017-08-31 23:47:21maltecreate