Enhancing NameNode Fault Tolerance in Hadoop Distributed File System

today's cloud computing environment, Hadoop is applied for handling huge data, tens of terabytes to petabytes, with commodity hardware (HDFS) for storage and software (MapReduce) for parallel data processing. In Hadoop version 1.0.3, there is a single metadata server called NameNode which stores the entire file system metadata in main memory and most of I/O operations are associated with those credential metadata. Hadoop is out of commission if NameNode is crashed because it works on memory which becomes exhausted due to multiple concurrent accesses (3). Therefore, NameNode is a single point of failure (SPOF) in Hadoop and it has to tolerate faults. To solve this issue, a proactive predictive solution is proposed for enhancing NameNode fault tolerance. The solution is designed to proactively calculate the predicted time to crash of NameNode due to resource exhaustion by evaluating the use of powerful Back Propagation Algorithm Neural Network. The proposed approach can give prediction accuracy with minimal error compared to the actual result. Therefore, NameNode's single point of failure can overcome through proposed proactively predicting the time to crash of NameNode caused by memory resource exhaustion.