Paper Title: 100x Faster Analytics from Reliable Computational Storage

Paper Abstract: The presentation will introduce a powerful solution that integrates massively parallel distributed data processing within an easily deployable software-defined computational storage framework ideal for modern flash-driven cloud environments. Network speed/bandwidth are inherent performance-limiting factors in any cloud architecture that separates elastic compute resources from storage services. While industry has developed a wide variety of omputational devices, they are restricted in their ability to provide application acceleration in the storage tier. AirMettle leverages the semi-structured nature of data itself to enable parallel processing on-demand using industry standard APIs. We will present AirMettle's architecture, which implements these innovations and techniques. Then we will demonstrate the relative performance of the solution-in-place analytics for data stored in CSV and JSON formats (both raw and client-side GZIP compressed) compared to the same results in AWS S3 using the same Select API. The additional benefits of using more optimized APIs will also be presented, along with discussions on how AirMettle benefits general database management systems as well.

Paper Author: Donpaul Stephens, CEO, AirMettle

Author Bio: Donpaul Stephens is the Founder & CEO of AirMettle, which has developed the first software-defined storage service which integrates true in-place parallel processing to enable magically fast response times for ad hoc queries on raw data. Donpaul is best known as the founder of Violin Memory where he developed the original concept; hired and led the core team; raised venture capital from strategic, institutional, and individual investors; established key strategic partnerships in both supply chain and go-to market partners; and developed multiple accounts. He has a BS in Computer Engineering and MS in Electrical & in Computer Engineering from Carnegie Mellon and an MBA from NYU.