1: \begin{abstract}

2: Solid state drives (SSDs) have seen wide deployment in mobiles, desktops,

3: and data centers due to their high I/O performance and low energy

4: consumption. As SSDs write data out-of-place, garbage collection (GC) is

5: required to erase and reclaim space with invalid data.  However, GC poses

6: additional writes that hinder the I/O performance, while SSD blocks can

7: only endure a finite number of erasures.  Thus, there is a

8: performance-durability tradeoff on the design space of GC.  To

9: characterize the optimal tradeoff, this paper formulates an analytical

10: model that explores the full optimal design space of any GC algorithm.  We

11: first present a stochastic Markov chain model that captures the I/O

12: dynamics of large-scale SSDs, and adapt the mean-field approach to derive

13: the asymptotic steady-state performance.  We further prove the model

14: convergence and generalize the model for all types of workload. Inspired

15: by this model, we propose a {\em randomized greedy algorithm (RGA)} that

16: can operate along the optimal tradeoff curve with a tunable parameter.

17: Using trace-driven simulation on DiskSim with SSD add-ons, we demonstrate

18: how RGA can be parameterized to realize the performance-durability

19: tradeoff.

20: \end{abstract}

21: