Design And Optimization Of Hash Algorithms Based On Multi-core CPUs

SIMD (Single Instruction Multi Data) instruction is an important part of the modernmicroprocessor instruction set, the SIMD instructions that are current widespread used isMMX instruction set and SSE instruction set in the x86architecture. In a SIMD instruction,multiple data parallel operation can be implement to achieve data-level parallelism andeffectively improve the efficiency of the implementation of the program. In addition, OpenMPis an application programming interface to write parallel programs on a multiprocessor, and aset of guidance notes used to multithreaded programs design on share memory parallel system.In multithread program design based on data diversity, OpenMP is a good choice. With theimprovement of a variety of password-protected technology, the Difficulty and computationare also increasing, it mainly behave in the increasing of one-way hash algorithm complexity.So, combining the SIMD and OpenMP parallel technology to redesign and optimization thehashing algorithm, will have important impact on information security as well ashigh-performance parallel computing.This paper focuses on the common MD4, MD5, SHA-1and RC4hash algorithm, as wellas some complex hash algorithms based on the basic algorithm, such as NTLM, MSCACHE,MYSQL5, UNIX_MD5and PDF file encryption algorithm. Firstly，this paper analysisprinciples of these algorithms, and then take advantage of the SIMD and OpenMP parallelcomputing power to achieve a the CPU parallel version of the algorithm. The various basichash algorithms encapsulated into the corresponding the CPU external interface, Functions inthe interface are support by SIMD, SSE2instruction set and OpenMP. Achieving complexhashing algorithm can directly call these external interface in parallel to calculate thecorresponding hash value. The experimental test result of SIMD parallel speedup is between3and4, the speedup of OpenMP is depend on total number of CPU threads, this results in highparallel efficiency of hash algorithms and prove that optimization strategies used in this paperare efficiency.