General speedup revisited

[connie]

See #78 for the older version of this issue.

I've noticed that on my computer, the minimal example running on 534cf3a takes 42 seconds vs the latest master at 21b83e3 takes 47 seconds, making it 5 seconds slower. I imagine this slowdown will affect larger models as well. @KEHANG presented at group meeting some changes which can help speed things up. I think there are also portions of code that could be cythonized or re-optimized. Please use this issue to brainstorm ideas or provide new profiling data.

[nickvandewiele]

#488 should make simulations with many edge reactions faster.

[nickvandewiele]

The figures shows the CPU times of make eg1 of the last 300 Travis CI builds triggered by ReactionMechanismGenerator/RMG-Py, i.e. pull requests into master and new branches.

[nickvandewiele]

Same for RAM used by make eg1

[rwest]

@walsh-p and @sherman-sc ran this job for about 7 hours before it crashed (an ODE solver problem). It got to 126 core species.
Here is the RMG.profile.pdf

Two remarkably large chunks:

• chemkin:1653:markDuplicateReaction 23.16% (23.17%) 135941×
• method 'update' of 'rmgpy.molecule.molecule.Molecule' objects> 27.61% (27.62%) 7599900×

Can we speed them up or call them less frequently?
Looks like we applied reaction recipes 2.5M times, so ran updateConnectivityValues 5M times (on average 2 new products per reaction?), but also ran update 7.5M times?

[mliu49]

Speed is still a problem, but this issue is well out-of-date.