Abstract:
The use of Java language for High Performance Computing (HPC) is becoming increasingly
popular due to appealing language features and availability of parallel programming libraries
and tools. In this work, we use MPJ (Message Passing Interface Java)-Express, a Java based
library to accelerate Advanced Encryption Standard (AES) algorithm. MPJ-Express is an MPI
(Message Passing Interface)-like implementation that supports acceleration of Java code on
multicore and cluster computer systems. We have partitioned the problem at two levels. By employing
a data parallel approach, we first divide the data length among available processors and
then data at each processor is further divided among processor cores. The experimental results
show almost linear throughput in case of multicore platform (1 node or stand alone system)
and non linear throughput for cluster platform. These experimental results are compared with
the AES algorithm accelerated by separately using other parallel programming tools in C language
such as OpenMP API (Open Multi Processing Application Program Interface), MPICH
(Message Passing Interface Chameleon) and CUDA (Compute Unified Device Architecture)
programming model. Parallel AES implementation using MPJ Express provides high speed up
factor and efficiency for multicore and cluster platform as compared to AES accelerated in C
using OpenMP and MPICH. But the speed up of GPU based implementation of AES in C using
CUDA (1 node) out performed AES in MPJ Express using multcore platform. Overall performance
of AES accelerated in C on all platforms is best as compared to AES accelerated using
MPJ Express. Accordingly it is concluded that this implementation is suitable for applications
that are platform independent.
