Analisis performance algoritme paralel block cipher RC6 menggunakan message passing interface (MPI)

Abstract

RC6 algorithm is a block cipher algorithm, which is one of the finalists in the AES competition. One of the RC6 algorithm parameters is the size of the block, where a larger block size is used the more time required to perform encryption-decryption. In this research the RC6 algorithm would be parallelized by dividing the data to each processor and the performance of the RC6 algorithm would be analyzed. Parallel implementation of the RC6 algorithm uses Message Passing Interface (MPI). The purpose of this research is to measure and analyze the performance of parallel RC6 algorithm. This research uses 4 computer with Linux operating system. This research uses files ASCII text and binary for the experiment. Each file will do the experiment three times repetition to take the average value. The results of this research show that execution time of parallel RC6 algorithm at the process of encryption and decryption is longer than execution time of sequential RC6 algorithm. For example, for file size of 249,149,712 bytes, encryption execution time of sequential RC6 algorithm is 17.91 seconds, whereas using the number of processors as many as 2, 4, 6 and 8, respectively are as follows 18.34 seconds, 20.37 seconds, 17.51 seconds and 18.93 seconds. The results of encryption execution time show that the encryption execution time by using the processor number of 4 is the slowest one. The speedup of parallel RC6 algorithm is less than one or smaller than the speedup of sequential RC6 algorithm. The speedup of parallel RC6 algorithm also not ideal because ideally speedup is equal to the number of processors used. The efficiency of parallel RC6 algorithm becomes lower along with more processor used. The resulting efficiency is far away from the optimum efficiency value, that is one, and it is not constant. To maintain the efficiency values if the number of processors is increased from p to p ' the input size should be increased in fraction of p'2 / p2. The cost of parallel RC6 algorithm increases as more processors are used. The resulting cost greater than the optimal cost value that is equal to Ts. The resulting overhead values become greater when the number of processors are increased. The results of this research indicate that the RC6 algorithm is not good when it is parallelized with domain decomposition method and using MPI. This is because the RC6 algorithm has a fast computational process, but it has the long communication time, so the total execution time of parallel RC6 algorithm longer than sequential RC6 algorithm.