In this paper, we develop a family of methods to characterize the behavior of new-generation Solid State Disks (SSDs). We first study how writes are handled inside the SSD by varying request size of writes
and detecting the placement of requested pages. We further examine how this SSD performs garbage collection and flushes write buffer. The result shows that the clustered pages must be written and erased simultaneously, otherwise significant storage waste will arise if such clustered pages are partially written.
We then conduct two case studies to analyze the storage efficiency when an SSD is used for server storage and the cache layer of a hybrid storage system. In the first case, we find that a moderate storage waste
exists, whereas in the second case, the number of written pages caused by a write request can be as much as 4.2 times that of pages requested, implying an extremely low storage efficiency. We further demonstrate
that most of such unnecessary writes can be avoided by simply delaying the issuance of internal write requests, which are generated when a read request cannot be serviced by the cache layer. We believe that this study is helpful to understand the SSD performance behavior for data-intensive applications in the big-data era.