The existence of computational intensive applications with a high degree of granularity which have to be executed as quickly as possible or sometimes with budget limitations makes the scheduling problem to be very important in today's distributed systems. In the context of the distributed systems being used by a various community of users, the quality and precision of the scheduling must be as close to optimality as possible. We present in this paper a scheduling method based on a genetic algorithm which respects the deadline constraints. Based on this model we developed a scheduler for MapReduce applications which also takes into account the heterogeneity of distributed systems. The developed scheduler was created for Hadoop platform, an already successful framework, and it can be built in conjunction with other existing Hadoop's schedulers like: fair-scheduler or capacity-scheduler. We tested and validated this model by testing its scalability and by comparing it with existing solutions in Hadoop. The novelty of this solution is that compared to other planners in Hadoop it meets the deadline and budget constraints and thus provides trust for the Hadoop platform.