The metrics introduced in SERVER-127796 do not exclude off-cpu time. This causes significant noise. On a busy system, then there is high CPU pressure, threads can spend a long time waiting to be scheduled. Currently this can accrue to unticketed aggregation metrics (as well as all the other interrupt and ticket-yielding delinquency metrics).

Our goal in the SERVER-127796 metrics is to measure the degree to which the query system is failing to cooperate with mongo's execution control / ticketing mechanisms. The off-CPU waits are totally unrelated.