Analytical modeling of garbage collection algorithms in hotness-aware flash-based solid state drives

Garbage collection plays a central role of flash-based solid state drive performance, in particular, its endurance. Analytical modeling is an indispensable instrument for design improvement as it demonstrates the relationship between SSD endurance, manifested as write amplification, and the algorithmic design variables, as well as workload characteristics. In this paper, we improve recent advances in using the mean field analysis as a tool for performance analysis and target hotness-aware flash management algorithms. We show that even under a generic workload model, the system dynamics can be captured by a system of ordinary differential equations, and the steady-state write amplification can be predicted for a variety of practical garbage collection algorithms, including the d-Choice algorithm. Furthermore, the analytical model is validated by a large collection of real and synthetic traces, and prediction errors against these simulations are shown to be within 5%.