Design and implementation of high-performance cloud computing system

Abstract

Cloud computing technology has brought huge changes to modern computing methods and is widely used. However, its security issues cannot be ignored. In order to use cloud services, users must store programs, data, etc. in the cloud. Once the attacker gains control of the server, or the cloud service provider itself is not trustworthy, the user's programs and data face great risks. For the security of cloud computing, SGX (Intel Software Guard Extensions) technology proposed by Intel provides users with a hardware-assisted trusted execution environment, and isolates the program code and data that need to be protected in a container called Enclave to ensure its confidentiality and completeness. However, SGX sacrifices some performance, and applications running in SGX cannot get satisfactory speed. Therefore, in order to improve the availability of the SGX application, it is necessary to optimize the SGX operating mechanism and program logic to increase the speed of the program. This article designs a set of high-performance cloud computing systems, selects a deep learning framework as a case, combines the characteristics of the framework and the SGX operating mechanism, analyzes various factors affecting performance, proposes an optimization scheme, and implements the system. Experiments show that the optimized trusted deep learning framework has reduced the running time by 24.8% compared with that before optimization.