Graph Based Workload Driven Partitioning System by using MongoDB

Abstract

The web applications and websites of the enterprises are accessed by a huge number of users with the expectation of reliability and high availability. Social networking sites are generating the data exponentially large amount of data. It is a challenging task to store data efficiently. SQL and NoSQL are mostly used to store data. As RDBMS cannot handle the unstructured data and huge volume of data, so NoSQL is better choice for web applications. Graph database is one of the efficient ways to store data in NoSQL. Graph database allows us to store data in the form of relation. In Graph representation each tuple is represented by node and the relationship is represented by edge. But, to handle the exponentially growth of data into a single server might decrease the performance and increases the response time. Data partitioning is a good choice to maintain a moderate performance even the workload increases. There are many data partitioning techniques like Range, Hash and Round robin but they are not efficient for the small transactions that access a less number of tuples. NoSQL data stores provide scalability and availability by using various partitioning methods. To access the Scalability, Graph partitioning is an efficient way that can be easily represent and process that data. To balance the load data are partitioned horizontally and allocate data across the geographical available data stores. If the partitions are not formed properly result becomes expensive distributed transactions in terms of response time. So the partitioning of the tuple should be based on relation. In proposed system, Schism technique is used for partitioning the Graph. Schism is a workload aware graph partitioning technique. After partitioning the related tuples should come into a single partition. The individual node from the graph is mapped to the unique partition. The overall aim of Graph partitioning is to maintain nodes onto different distributed partition so that related data come onto the same cluster.