Abstract: MapReduce programming model enjoys widespread application and becomes new popular parallel programming paradigm nowadays. It uses MapReduce model to parallelize coarse-grained genetic algorithms,and takes mull objective optimization problem as the benchmark. All the experiments are made under hadoop and a cluster which consists of commodity servers. When the number of variable reach to 10E}7，the efficiency of parallel algorithm can be multiplied several times without introducing any bottlenecks revolved memory. Finally we studied how the parallel degree af- feels the performance.

Key words: Keywords Largcscale variables, MapReduce, Parallel genetic algorithms, Mult objective optimization, Performance analysis