Designing fast and scalable algorithm for mining frequent itemsets is always being a most eminent and promising problem of data mining. Apriori is one of the most broadly used and popular algorithm of frequent itemset mining. Designing efficient algorithms on MapReduce framework to process and analyze big datasets is contemporary research nowadays. In this paper, we have focused on the performance of MapReduce based Apriori on homogeneous as well as on heterogeneous Hadoop cluster. We have investigated a number of factors that significantly affects the execution time of MapReduce based Apriori running on homogeneous and heterogeneous Hadoop Cluster. Factors are specific to both algorithmic and non-algorithmic improvements. Considered factors specific to algorithmic improvements are filtered transactions and data structures. Experimental results show that how an appropriate data structure and filtered transactions technique drastically reduce the execution time. The non-algorithmic factors include speculative execution, nodes with poor performance, data locality & distribution of data blocks, and parallelism control with input split size. We have applied strategies against these factors and fine tuned the relevant parameters in our particular application. Experimental results show that if cluster specific parameters are taken care of then there is a significant reduction in execution time. Also we have discussed the issues regarding MapReduce implementation of Apriori which may significantly influence the performance.