Additional comments:

  * Why not use replication in combination with migration?
  * Rationale of the model overall is not clear.
  * Cello99 workload is shown in Table 2, but not discussed in the paper.
  * How the data is finally partitioned on both devices is not clear.
  * Equaling lifetime of HDD and SSD is not logical. Why you want to make HDD last as long as SSD since the former is much costly?