Le 08/06/2014 22:02, Clément Bera a écrit : > Some collection protocols were optimized so your result makes sense.
The profiles show it is faster on 1.4! > The current performance bottleneck for your application definitely comes > from the use of isKindOf: based on your profiles. inheritsFrom: is > called by isKindOf: I guess. Although there are use of iskindOf: in the image (384), there are exactly 0 use case from DrGeo source code directly, but still impacted I guess. The situation is not different from 1.4, only inheritsFrom: seems a lot slower in 3.0. What is strange, and may give some indication about what is happening underneath, the script as execution goes on, is slower and slower under Pharo 3.0. With Pharo 1.4, the execution speed is completely constant (as perceived). > Now be very careful the current GC has very strange side effect. Did you > perform manually a full GC before each bench ? Ok. I added at the begining of the script a garbage collection. The result are more or less the same. Now I wonder could it be related to collection during execution: **Memory** old +461,292 bytes young -1,436,444 bytes used -975,152 bytes free +975,152 bytes **GCs** full 0 totalling 0ms (0.0% uptime) incr 48 totalling 213ms (19.0% uptime), avg 4.0ms tenures 8 (avg 6 GCs/tenure) root table 0 overflows Hil