Revisiting Operating System Mass Storage Presumptions Enables Higher Performance and Efficiency

Random access mass storage has been a pillar of computing since the 1956 delivery of the first IBM 350. The interface paradigm of position then transfer is pervasive, despite exponential advances in access speeds, transfer performance, density, and reliability.Consistency in interface paradigm does not mean that implementations have remained static, there have been numerous improvements: multiple heads, incremental positioning, recording technologies, and storage caches. While multiple levels of storage caches have become ubiquitous, caches are not a panacea. Caches benefit physically sequential usage and repetitive use by reducing physical media accesses. However, systems with hundreds or thousands of actively accessed files can clog caches with data that is only used a single time, yielding cache pollution.Advances in semiconductors have vastly increased processing power and memory capacity, particularly since 1990. Solid state mass storage eliminates rotational delay, but can still have other implementation-related delays. Hierarchical mass storage provides the illusion of transparency, but as with virtual memory, delays attributable to data migration between levels cannot be concealed.Through the 1980s, resource limitations obliged I/O infrastructure implementors to choose minimum resource implementations; other choices were infeasible. Feasibility is mandatory; efficiency and performance are desirable. Better resourced environments allow higher efficiency and performance with some increase in processing and memory consumption.We will examine an approach that preserves the pre-existing I/O API, with minimal culturally-compatible extensions that enable higher performance and increased efficiency.