I think I've finnaly (
) figured the root cause for eMF long drops, so a brief history review :
beta 1 : long lasting drops
beta 2 : adding slots based on amount of fully activated slots and used upload speed (originally commented out on eMF) - long lasting drops at reduced rate
beta 4 : zz bandwidth control - drops are gone, however due to the "slot injection" too mant slot on high upload bandwidth
There is (to my humble opinion) a serious bug in eMF trickle calculation : it is based on current upload rate instead of on upload limit / USS. The result was that if not enough slot were fully activated number of slots remained low (very few trickles were desired since bandwidth was not utilized). My original change (beta 2 - added slots) was faulty.
I'm now finishing sanity checks with fixed code (added slots from beta 2 - gone, trickel slots number determined by UL limit or USS instaed of current upload speed) - probably a new beta tommorow (AcK as well)