Classifying SSD input-output streams to optimize storage overprovisioning

The data update mechanism for a solid-state drive (SSD) involves overprovisioning, e.g., a storage footprint larger than the actual data size, and write amplification, e.g., a physical amount of information written being a multiple of the actual amount of information. Both overprovisioning and write amplification overheads are at least partially a result of treating input data streams as statistically indistinguishable streams and subjecting them to fixed, pre-configured provisioning. This disclosure describes techniques to classify the input-output streams of an SSD into various types, e.g., non-overlapped sequential data, chunked data streams of variable length, etc. Overprovisioning is optimized for each input stream based on its classification. The techniques thereby reduce both overprovisioning and write amplification overheads, resulting in SSD life and throughput that is significantly better than conventional techniques