Update: I just conducted a test using Good After Time / Good Till Time orders. The idea is that since the processing is all done on IB's side, the variable of my own ping to IB is eliminated.
My methodology is as follows:
1) Choose 100 stocks with wide spreads in postmarket.
2) Submit orders 1c inside the spread, Good After Time X, Good Till Time X + 10 seconds.
3) Look for my quotes on the tape, and note when the price changes (Y) and changes back (Z)
4) Calculate statistics for Submit (Y-X) and Cancel (Z-(X+10))
results (all numbers in milliseconds):
Submit: min: 26.0, median: 123.0, mean: 147.9, max: 384.0, std. dev 119.8
Cancel: min: 1.0, median: 2.0, mean: 1.9, max: 3.0, stddev: 0.7
Clearly canceling is much, much faster than submitting. The fact that submitting is slower on average than what other people noted (for regular orders instead of Good After Time) in this thread is suspicious, possibly indicating that the Good After Time system goes through a different, slower path.
In the future I plan to test submitting from my own system to see what I can get. I may also contact IB to ask why Good After Time is so slow, IMO a mean 147ms / stddev 119ms for submit is not acceptable.
