In two ways:
(1) When there's a session timeout, restart the measurement at a later time
instead of just stalling.
(2) When there's a port change, do not re-measure the interval if the current
interval works well in the past. We keep success count and decrement it
by half when there's a port change. When the count is below a threshold,
we restart the measurement process.
Change-Id: I7256464435a5e2d2a239bfccaa004e9ceb1d9ce5
129d0b08fd