The Cycle of Adjustment
At the heart of nearly all stable systems — from your home thermostat to complex cloud autoscalers — is the Feedback Loop. It is a continuous cycle of four steps:
- Measure: Observe the actual current state of the system (e.g., "Current CPU is at 80%"). This is called the Process Variable (PV).
- Compare: Calculate the difference between the actual state and the desired state. This difference is called the Error. (e.g., "Target is 50%, Current is 80%. Error = +30%"). The desired state is called the Setpoint (SP).
- Decide (The Controller): Determine what action to take based on the size and direction of the error. (e.g., "Error is large and positive, we need to add capacity"). The controller is the algorithm that translates error into action.
- Act: Apply the change to the system (e.g., boot up two new VM instances). This output is called the Control Variable (CV) or Manipulated Variable.
The cycle then immediately repeats. The system is always watching itself and correcting. The effectiveness of your system depends entirely on how smart your Controller (Step 3) is — a naive controller can make instability worse, while a well-tuned one can make a chaotic system rock-solid.
This type of loop is called a closed-loop or negative feedback system. "Negative" here doesn't mean bad — it means the feedback opposes the error (it pushes back against deviation). A thermostat that turned the heat up when the room was already too hot would be a positive feedback loop, and it would be catastrophic.