MapReduce Programming Model and its role in Hadoop.

In the Hadoop framework, MapReduce is the programming model. MapReduce utilizes the map and reduce strategy for the analysis of data. In today’s fast-paced world, there is a huge number of data available, and processing this extensive data is one of the critical tasks to do so. However, the MapReduce programming model can be the solution for processing extensive data while maintaining both speed and efficiency. Understanding this programming model, its components, and execution workflow in the Hadoop framework will be helpful to gain valuable insights.

What is MapReduce?

MapReduce is a parallel, distributed programming model in the Hadoop framework that can be used to access the extensive data stored in the Hadoop Distributed File System (HDFS). The Hadoop is capable of running the MapReduce program written in various languages such as Java, Ruby, and Python. One of the beneficial factors that MapReduce aids is that MapReduce programs are inherently parallel, making the very large scale easier for data analysis.

When the MapReduce programs run in parallel, it speeds up the process. The process of running MapReduce programs is explained below.

• Dividing the input into fixed-size chunks: Initially, it divides the work into equal-sized pieces. When the file size varies, dividing the work into equal-sized pieces isn’t the straightforward method to follow, because some processes will finish much earlier than others while some may take a very long run to complete their work. So one of the better approaches is that one that requires more work is said to split the input into fixed-size chunks and assign each chunk to a process.
• Combining the results: Combining results from independent processes is a crucial task in MapReduce programming because it may often need additional processing such as aggregating and finalizing the results.

Key components of MapReduce

There are two key components in the MapReduce. The MapReduce consists of two primary phases such as the map phase and the reduces phase. Each phase contains the key-value pairs as its input and output and it also has the map function and reducer function within it.

• Mapper: Mapper is the first phase of the MapReduce. The Mapper is responsible for processing each input record and the key-value pairs are generated by the InputSplit and RecordReader. Where these key-value pairs can be completely different from the input pair. The MapReduce output holds the collection of all these key-value pairs.
• Reducer: The reducer phase is the second phase of the MapReduce. It is responsible for processing the output of the mapper. Once it completes processing the output of the mapper, the reducer now generates a new set of output that can be stored in HDFS as the final output data.

Execution workflow of MapReduce

Now let’s understand how the MapReduce Job execution works and what all the components it contains. Generally, MapReduce processes the data in different phases with the help of its different components. Take a look at the below figure which illustrates the steps of the job execution workflow of MapReduce in Hadoop.

• Input Files: The data for MapReduce tasks are present in the input files. These input files reside in HDFS. The format for input files is arbitrary, while line-based log files and binary format can also be used.
• InputFormat: The InputFormat is used to define how the input files are split and read. It selects the files or objects that are used for input. In general, the InputFormat is used to create the Input Split.
• Record Reader: The RecordReader can communicate with the Input Split in the Hadoop MapReduce. It can also convert the data into key-value pairs so that the mapper can read. By default, the RecordReader utilizes the TextInputFormat for converting data into key-value pairs. The Record Reader communicates with the Input Split until the file reading is completed. It then assigns a byte offset (unique number) to each line present in the file. Then these key-value pairs are sent to the mapper for further processing.
• Mapper: From the RecordReader, the mapper receives the input records. The Mapper is responsible for processing those input records from the RecordReader and it generates the new key-value pair. The Key-value pair generated by the mapper can be completely different from the input pair. The output of the mapper which is intermediate output is said to be stored in the local disk since it is the temporary data.
• Combiner: The Combiner in MapReduce is also known as Mini-reducer. The Hadoop MapReduce combiner performs the local aggregation on the mapper’s output which minimizes the data transfer between the mapper and reducer. Once the Combiner completes its process, the output of the combiner is passed to the partitioner for further work.
• Partitioner: In Hadoop MapReduce, the partitioner is used when we are working with more than one reducer. The Partitioner extracts the output from the combiner and then it performs partitioning. The partitioning of output takes place based on the key and then it is sorted. With the help of a hash function, the key (or subset of the key) is used to derive the partition. Since MapReduce execution works with the help of key-value, each combiner output is partitioned and a record having the same key value moves into the same partition and then each partition is sent to the reducer. Partitioning of the output of the combiner allows the even distribution of the map output over the reducer.
• Shuffling and Sorting: The Shuffling performs the shuffling operation on the mapper’s output before it is sent to the reducer phase. Once all the mapper has completed their work and their output is said to be shuffled on the reducer nodes, then this intermediate output is merged and sorted. This sorted output is passed as input to the reducer phase.
• Reducer: It takes the set of intermediate key-value pairs from the mapper as the input and then it runs the reducer function on each of the key-value pairs to generate the output. This output of the reducer phase is the final output and it is stored in the HDFS.
• Record Writer: The Record Writer holds the power of writing these output key-value pairs from the reducer phase to the output files.
• Output format: The Output format determines how these output values are written in the output files by the record reader. The Output format instances provided by Hadoop are generally used to write files on either HDFS or on the local disk. Thus, the final output of the reducer is written on the HDFS by output format instances.

Conclusion

We have seen the concept of MapReduce and it's components in the Hadoop framework. Understanding these key components and the MapReduce workflow provides a comprehensive insight into how Hadoop leverages MapReduce to solve complex data processing challenges while maintaining both speed and scalability in the ever-growing world of big data.