Minfs micro-benchmarks are targetted to keep check on a few aspects of the filesystem performance. The suite is designed to catch regressions and to help one understand the impact of their changes at micro level.
The basic principle is
For example, if we know device properties, mkfs and mount options, then we can tell with certainty what blocks mount should read and write to. Similarly, if there were no operations performed after mount, we can tell what IOs an unmount should perform. Using such building blocks we should be able to understand and keep track of IO stats for more complex FS operations like write, create and delete, etc.
Since we want our accounting to check those of filesystem, we intend to maintain our own accounting routines for FS operations. These routines, like AddMountCost, will rely on the state (like clean/dirty) of the filesystem. These routines compute their number from first principles and their definitions can be found in *-costs.cc files.