Empowering Data Centers with Computational Storage Drive-Based Deep Learning Inference Functionality to Combat Ransomware

Document Type

Conference Proceeding

Publication Date

1-1-2024

Abstract

The exponential increase in data has significantly burdened data centers with heightened operational demands and challenges in managing, storing, and processing information. Computational Storage Drives (CSDs) mitigate these issues by accelerating processes, reducing energy consumption, and liberating CPU resources for other data center tasks. In light of the notable increase in harnessing the pattern recognition capabilities of Machine Learning (ML) for a variety of data center applications coupled with the aforementioned advantages of leveraging CSDs in data center environments, we build upon the state-of-the-art in CSD development via proposing a unique method for offloading ML inference to CSDs, such as Samsung's SmartSSD. Using various parallelization and optimization techniques, we demonstrate that the proposed CSD-based ML inference method surpasses the inference speed of a high-performance GPU by 344.6×. We further showcase the ability of the presented approach to promptly detect ransomware with high accuracy, precision, recall, and F1 scores, enabling effective and timely mitigation directly within the CSD. Indeed, this ML inference strategy offers the potential to enhance an assortment of other data center tasks.

Publication Source (Journal or Book title)

Proceedings - 2024 54th Annual IEEE/IFIP International Conference on Dependable Systems and Networks - Supplemental Volume, DSN-S 2024

First Page

125

Last Page

132

This document is currently not available here.

Share

COinS