Self-Configuration of the Number of Concurrently Running MapReduce Jobs in a Hadoop Cluster

There is a trade-off between the number of concurrently running MapReduce jobs and their corresponding map and reduce tasks within a node in a Hadoop cluster. Leaving this trade-off statically configured to a single value can significantly reduce job response times leaving only sub optimal resource usage. To overcome this problem, we propose a feedback control loop based approach that dynamically adjusts the Hadoop resource manager configuration based on the current state of the cluster. The preliminary assessment based on workloads synthesized from real-world traces shows that the system performance can be improved by about 30% compared to default Hadoop setup.