Abstract
In this paper, the schedulers of MapReduce on the Hadoop platform are analyzed. LATE (Longest Approximate Time to End) scheduling algorithm always launches backup tasks for those tasks which have more remaining time than other tasks. This algorithm redistributes the backward task to fast nodes and does not consider fast node workload, which may result in cluster performance degradation. Aiming at its disadvantages, an improved LATE scheduling algorithm is put forward. The algorithm employs a method of CPU occupancy as a load index by feeding back the load index and adapted load changes dynamically. The experiment results on Hadoop cloud platform show that the improved LATE scheduling algorithm can shorten the mission completion time and improve the whole performance of the cluster.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
