Exploring a Software-Hardware Framework for Computational Storage

Computational Storage platforms are emerging as a disruptive enhancement in the data processing ecosystem as they have the potential to address the critical bottleneck of limited I/O bandwidth and high latency between storage and compute nodes in the system hierarchy. This “late coming” technology to the Exascale program has the potential to enhance DOE and the broader government’s leadership in Exascale computing, AI, and data science. Commercially, the applicability of a new storage platform to data science cannot be underestimated, as data is frequently stored with the idea that it will be useful once the computational capability exists to extract meaningful insights. This project facilitates the enablement of computational storage platforms by rapidly extending existing storage interfaces with computational storage capability, by creating a new storage system architecture that enables richer utilization of that capability, and by pursuing a fully commodity approach to computational storage. It supports the disruptive potential to “unlock data” in both data-driven science and AI/Machine Learning.