Figure 3

Typical load during pre-computation on the Google App Engine instances during partial bulk addition of the MovieLens 100K dataset. Note that even if the latency, for example, in the third row is only 175.3ms, it does not equate to a QPS value of $\frac{1}{175.3}=0.0057$. The QPS signifies the number of queries the particular instance receives on an average while the latency signifies how long the instance takes on an average to process one such query. The fact that the latency is low does not imply that the QPS will be high because the QPS is the result of the queries sent to the front-end, the speed of which is affected by external factors, such as network latency, GAE/J queue processing speed and so on.