While the industry has been focusing on improving the IOPS & density of storage, the growth rate hasn’t been the same for both across HDD’s, Flash & SCM (storage class memory). In order to keep up with the required IOPS/TB, applications pay the cost with the latency credits. Using a RocksDB use-case, we discuss about the various considerations & trade-offs in the HW/SW co-design that influences an optimal solution that solves for predictable IO latency.
