Forum Discussion
Hamish
Apr 11, 2010Cirrocumulus
The only one of the stats you're looking at that are useful for SSL TPS is the (total) number of connections. From that, if you poll at regular intervals, you can get the average rate over the poll time. But since the licensing works on a 10ms interval, you'd have to poll at 5ms at least in order to get your stats fine grained enough to work out how close you were to the limit (On the basis that you have to sample at 2x the required resolution and the required resolution is 10ms).
I don't believe that command line stats are fine grained enough... What's really required is a high resolution count of SSL connections... Something you can poll every second and get 100 entries for the last seconds 10ms intervals.
What would be really useful is if the licensing 'borrowed' connections from other 10ms slots. It SHOULD (To be fair) be counting TPS over the LAST second... Not 1/100th over the last ms... After all, just because the last 10ms were unusually busy DOES NOT imply that the other 99x 10ms slots were also as busy.
I'll stop here, because I have issues with the way this is done when you buy 1000TPS, I expect 1000TPS. Not 10TP/10ms slot... HTTP is not constant rates...
pragma almost_rant_mode off...
H