HaSTE: Hadoop YARN Scheduling Based on Task-Dependency and Resource-Demand

摘要

The MapReduce framework has become the de facto scheme for scalable semi-structured and un-structured data processing in recent years. The Hadoop ecosystem has evolved into its second generation, Hadoop YARN, which adopts fine-grained resource management schemes for job scheduling. One of the primary performance concerns in YARN is how to minimize the total completion length, i.e., makespan, of a set of MapReduce jobs. However, the precedence constraint or fairness constraint in current widely used scheduling policies in YARN, such as FIFO and Fair, can both lead to inefficient resource allocation in the Hadoop YARN cluster. They also omit the dependency between tasks which is crucial for the efficiency of resource utilization. We thus propose a new YARN scheduler, named HaSTE, which can effectively reduce the makespan of MapReduce jobs in YARN by leveraging the information of requested resources, resource capacities, and dependency between tasks. We implemented HaSTE as a pluggable scheduler in the most recent version of Hadoop YARN, and evaluated it with classic MapReduce benchmarks. The experimental results demonstrate that our YARN scheduler effectively reduces the makespans and improves resource utilization compare to the current scheduling policies.