Nice that you get it working 
!
Measuring on the JVM is not that straightforward, you need to warmup the JVM first and take a few hundred routes, for CH I use even 5K. E.g. have a look into Measurement.java which can be started easily via ./graphhopper.sh measurement your.pbf
Regarding the connection not found → see this discussion as it could be related: 0.5.0: Connection between locations not found
but I’ve now realised I should have left it in (and just returned the square of the planar distance).
Not sure how much that’s slowed it down, but I doubt it’s too much.
One always has to test this, those sometimes confirms my guesses but sometimes I was already proven wrong … the JVM does sometimes tricky optimizations which one cannot guess.
and if I haven’t done something silly to give misleading results, then it looks pretty good for CH.
Especially as it’s essentially a net gain (both in terms of speed / simplicity) for “free”
Yes, indeed nice.
All I really added was a new DistanceCalc for planar coordinates.
Ok, hopefully I have not worked around it somewhere …
Finally, I’m assuming most of the speedup occurs in the location lookups.
The location lookup is also measured in ‘measurement’ separately