The track misregistration (TMR) requirement for hard disk drives (HDDs) has become more demanding to keep pace with increase in recording density. It is no longer sufficient to make design decisions for actuator, servo, power amplifier and formatting efficiencies in isolation. Inclusion of these factors into servo performance criteria makes the servo optimization non-convex and non-linear. However, the structure of the track following servo controller for HDDs is usually simple for computational efficiency. This simplicity makes optimization in the face of non-linear performance criteria numerically tractable. We demonstrate the effectiveness of random neighborhood search algorithms in achieving servo design optimization. We emphasize optimization techniques that are useful with analytical models (available commonly during the research/advanced development stage) as well as numerical data (available during product development and manufacturing).