Название: On Architecting Fully Homomorphic Encryption-based Computing Systems
Автор: Rashmi Agrawal, Ajay Joshi
Издательство: Springer
Серия: Synthesis Lectures on Computer Architecture
Год: 2023
Язык: английский
Формат: pdf (true), epub
Размер: 10.2 MB

Discusses design challenges for an FHE-based system, using different hardware backends (CPU, GPU, FPGA, and ASIC).

This book provides an introduction to the key concepts of Fully Homomorphic Encryption (FHE)-based computing, and discusses the challenges associated with architecting FHE-based computing systems. Readers will see that due to FHE’s ability to compute on encrypted data, it is a promising solution to address privacy concerns arising from cloud-based services commonly used for a variety of applications including healthcare, financial, transportation, and weather forecasting. This book explains the fundamentals of the FHE operations and then presents an architectural analysis of the FHE-based computing. The authors also highlight challenges associated with accelerating FHE on various commodity platforms and argue that the FPGA platform provides a sweet spot in making privacy-preserving computing plausible.

This book is organized as follows:

• Chapter 1 introduces the notion of privacy-preserving computing (PPC) along with a brief description of various cryptographic and non-cryptographic PPC techniques. This chapter also introduces the FHE-based PPC technique, and provides its history and background.

• As the focus of this book is on CKKS FHE scheme, Chapter 2 describes in detail all the primitive operations in the CKKS FHE scheme by explicitly writing out algorithms for each of these operations. Chapter 2 also describes the bootstrapping operation, and then presents a complete FHE-based computing example.

• Chapter 3 first describes the key performance bottlenecks in the CKKS FHE scheme and then presents the profiling results (based on arithmetic intensity analysis) of individual primitive operations. This chapter also describes the bootstrapping operation that has been designed using these primitive operations followed by its arithmetic intensity analysis. Finally, this chapter discusses the arithmetic intensity analysis of an example CKKS-based logistic regression training application. Based on various arithmetic intensity analysis, this chapter concludes that the bootstrapping operation and its encompassing CKKS-based applications are memory bound.

• Chapter 4 presents an analysis of CKKS-based computing on example CPU, GPU, ASIC, and FPGA. This analysis reveals that when executing CKKS-based applications, both CPUs and GPUs are limited in performance because of the limited main memory bandwidth, ASICs outperform CPUs and GPUs but are expensive, and FPGAs provide an affordable and practical solution. This chapter describes several trade-offs that we need to consider when using each of these compute platforms for CKKS-based computing.

• Chapter 5 concludes the book with a description of some of the emerging architectures and approaches that could be used for accelerating FHE-based applications.




ОТСУТСТВУЕТ ССЫЛКА/ НЕ РАБОЧАЯ ССЫЛКА ЕСТЬ РЕШЕНИЕ, ПИШИМ СЮДА!