Paper Title: Computational Storage for the Masses

Paper Abstract: Computational storage is an emerging industry initiative that allows a storage drive to offload targeted functions from a host application, thereby significantly improving performance as well as TCO benefits. To date, computational storage has been targeted for specific and limited applications, curtailing broad adoption. In this presentation, we will provide an example of an SSD controller solution that will be based on the NVMe TP4091 definition. The goal is to provide a native Linux-based framework, allowing an end-user to download any desired functionality, provided by a third party or internal developed by the customer. In doing so, any user can select which function that requires acceleration to be run within the SSD controller.

Paper Author: Mats Oberg, Sr. Director, DSP, Marvell

Author Bio: Mats Öberg is Sr. Director DSP, Storage Office of the CTO at Marvell. He leads research and development of signal processing for SSDs and HDDs. He also is responsible for computational storage and storage related machine learning. Since joining Marvell, Mats has been leading development of read channels for perpendicular recording and TDMR, as well optical recording for DVD and Bluray. Mats earned his PhD in Electrical Engineering from University of California, San Diego.