It's a complicated question:
- on linux, there is hardly a difference between running with sumo-gui and delay 0 versus sumo
- on windows, delay 0 introduces more delay than it should most likely due to some event handling overhead
- on windows, setting delay 10 has a stronger slowdown effect than on linux
- nothing was changed between 1.15 and 1.17 that should have changed this
- bugfixes in sumo itself can lead to changing scenario behavior and even more/less jamming which affects the number of vehicles at any given time and thus simulation speed
So to figure this out:
- is the speed difference maybe from sumo itself? is there a speed difference in your scenario when running without gui using 1.15 and 1.17?
- did the number of UPS (updates per second reported in the verbose log) change when running your scenario with either version (again without gui)?
- what OS are you running your tests on?
- what are the running times for sumo vs sumo-gui with delay 0 and what is the reported real time factor and UPS respectively?