Explore a groundbreaking 19-minute conference talk from LCTES 2023 that introduces ISVABI, an innovative In-Storage Video Analytics Engine with Block Interface. Discover how this software/firmware approach maintains the SSD block I/O interface while providing offloaded operations for user-space video analytics workloads without requiring SSD hardware modification. Learn about the implementation on the Cosmos+ OpenSSD platform and its impressive performance improvements, outperforming normal SSDs by 4.18x for various video operations while consuming 16% less power. Gain insights into the evaluation of ISVABI on five real-world video analytics workloads, demonstrating a 1.89x end-to-end latency improvement. Understand the challenges of processing large volumes of video data and how ISVABI addresses these issues without changing data structures within SSDs or requiring re-programming of existing applications. Presented by researchers from Pennsylvania State University, University of Virginia, and Intel Labs, this talk offers valuable knowledge for those interested in advanced video analytics, SSD technology, and efficient data processing solutions.
Overview
Syllabus
[LCTES'23] ISVABI: In-Storage Video Analytics Engine with Block Interface
Taught by
ACM SIGPLAN