In order to understand the insight effect of WOLF, we compare our design with the most recent LFS using adaptive method which is the baseline system. The reason is that we want to explore the effect with our reordering write buffers rather than the adaptive cleaning policy. Therefore, two compared systems use the same adaptive garbage cleaning strategy. In experiments, system automatically selects either benefit-to-cost or the hole-plugging depending on the cost-benefit estimates. WOLF separates active data from cold data to generate active/inactive segments in initial writes. The different disk layouts in two systems lead to different performance.
In our experiments of this paper, we set several default parameters unless specified: a 64 MB read cache, each disk segment is 256 KB and each segment buffer is 256 KB.