Performance Comparison of State of Art NoSql Technologies Using Apache Spark

Data is the new currency of digital world today. Data generated in last 2 years are more in size as compared to data generated in last 15 years. The nature of data generated have varying dimensions, size, speed and behavior along with being semi and full unstructured, it also contains various formats including text, document, excel, power point, web blogs, posts, chats, tweets, audio and video streams and long range numeric values, etc. Storing such type of data in legacy SQL based storage will not yield the benefit of currency. To take full advantage of data the IT industry is equipped with variety of State of Art NoSql (Not only Sql) databases. Each of them has their own specific features and limitations. In this research we have conducted an experiment on state of art NoSql technologies to find out a comparative analysis among them on the basis of performance, integration, ease of use and size of data loading/unloading capabilities. For experiment we used 3.4 TB of data which contains medical test records, lab diagnostics and prescriptions, long range pi values. The generated data was stored in AeroSpike, BerkeleyDB, CouchBase, HBase, MongoDB and Redis. The performance testing was done on queries like search in, equate, greater than, less than and other general arithmetic operations, etc. Those queries were executed using the Apache Spark on a cluster with a processing capacity of 54 cores and memory of 168 GB. The comparison provided some useful and defining results towards selection of NoSql stores for specific nature of jobs.