He's right. You guys need to see the system in terms of it's derivative. The rate of speed in which data is produced and consumed. This article is saying you b

He's right. You guys need to see the system in terms of it's derivative. The rate of speed in which data is produced and consumed. This article is saying you basically want an average negative velocity where consumers eat data faster then producers on average. Any queue that is non empty means the velocity is positive. Producers are making data faster then consumers and you have a limited time window before the queue becomes full and your system starts dropping data. If you observe that your system never has a full queue but is constantly loaded with partially filled queues it means you built and tuned your system such that it operates on the border of failure. It means your system is continuously hitting unsustainable velocities all the time, but stops short of total failure (a full queue). Remember, think about the system in terms of it's derivative. The average derivative must be negative for your system to even function. Any time a queue is filled partially or full it means there was a period of positive velocity. These periods cannot exceed the amount of time the system is in negative velocity.