That's interesting, thanks. I am curious, where are you measuring the end of the 250ms, is that when the execution happens on the exchange, or is it when you get the execution acknowledgement back through the API?
Also, you said "fills completely", does that mean it fills partially much faster...