Yarn contrast MapReduce1

1. Scalability: In contrast to Jobtracker, each application instance, here can be said to be a mapreduce job has a managed application management that runs during application execution. This model is closer to the original Google paper.
2. High availability: Highly available (high availability) usually after a service process fails, another daemon (daemon) can replicate the state and take over the work. However, for a large number of rapidly complex state changes, in jobtracker memory, making it very difficult to guarantee HA for this service, by splitting the responsibility on Jobtracker, a resource manager and a application Master is in yarn. Making a usable change a divide-and-conquer problem.
3. Utilization: MapReduce1 Each tasktracker is configured with a statically assigned fixed-size slots. Assign it as map slots and reduce slots when configured, a map slot can only run the map task, and a reduce slot can only run the reduce task. In yarn, a Node Manager manages a resource pool, and in MapReduce 1, reduce has to wait until the map task runs out.
4. Multi-tenancy (multitenancy): The biggest benefit of yarn is that it opens the door to Hadoop for other distributed applications. MapReduce is just one of those applications.

Yarn contrast MapReduce1