Eliminating the sawtooth

[richb-hanover]

I have always been surprised the sawtooth display that appears in latency charts. Here's an example:

It doesn't match my intuition of the physical effects that underly the measurements. I can see why the front edge is so sharp - there was a big increase. But why does the back edge always slope down? And why is it always so linear?

I think I figured it out, by watching this latency chart:

The spike began at -1.11 seconds, and it was back to (near) zero at -1.03 seconds. That's a difference of 80msec, and lo and behold, the spike was 80 msec tall. Ahah! And seems to match the other observations that I've seen.

My conclusion is that Crusader is "deaf" during the time that it's waiting for a response to the UDP ping. When it finally returns (say, 80msec later), Crusader records the ping time. If the next UDP ping returns in a couple msec, the line connecting the two samples shows that linear, downward slope.

This means the plot doesn't correctly represent what's really happening.

Am I guessing the correct mechanism? Thanks.