TR-2020-02
Empowering Architects and Designers: A Classification of What Functions to Accelerate in Storage
Chen Zou; Andrew Chien. 2 June, 2020.
Communicated by Andrew Chien.
Abstract
Storage acceleration is a topic of widespread study and innovation. We seek fundamental understanding of 1) what computations to offload to benefit application performance and scalability and 2) how the properties of such computations shape offloading benefits. Using raw-data analysis as an exemplar, we study acceleration benefits using SparkSQL operating in a cloud data center, analyzing where and how speedup and scalability are improved.From this base, we analyze 14 function offload candidates drawn from 17 research studies to create a classification to help the application or hardware system designer. We analyze from the perspective of feasible acceleration, network data reduction, and software change - for application and system software. These studies identify functions that are clear wins and should be the target of early “computational storage”, those that give clear benefit, but face software challenges to support, and finally those that are less compelling.
Original Document
The original document is available in PDF (uploaded 2 June, 2020 by Andrew Chien).