摘要：Copy-On-Write (COW) is a powerful technique for data protection in file systems. Unfortunately, it introduces a recursively updating problem, which leads to a side effect of write amplification. Studying the behaviors of write amplification is important for designing, choosing and optimizing the next generation file systems. However, there are many difficulties for evaluation due to the complexity of file systems. To solve this problem, we proposed a typical COW file system model based on BTRFS, verified its correctness through carefully designed experiments. By analyzing this model, we found that write amplification is greatly affected by the distributions of files being accessed, which varies from 1.1x to 4.2x. We further found that write amplification is also affected by the number of files being accessed, the number of files contained in a file system, and as well as the space utilization of file system trees.
