Results are presented of a performance comparison between SQL Server and NoSQL MongoDB database systems. The comparison involved insert, three kinds of update, four kinds of simple select, and three kinds of complex select operations. The data used was structured and concerned departments, users, and projects. Four test cases were each run 100 times. The number of tuples used ranged from 148 in the first test case to 12,416 in the fourth test case. The authors sought to answer the question of whether or not MongoDB was a viable solution for a modest-sized database with structured data.

Figures 3 to 13 clearly show that MongoDB equaled or outperformed SQL Server in most cases. The authors believe MongoDB’s performance advantage is attributable to its indexing mechanism and its use of memory-mapped files. However, for the complex select operation involving the computation of average, MongoDB was almost 23 times slower than SQL Server. The lack of built-in aggregation functions in MongoDB required the writing and use of a MapReduce method, and this greatly slowed things down. The authors conclude that MongoDB could be a good solution where a database schema changes a lot with time and queries do not involve aggregation functions.

While there is no discussion of benchmark approaches, at least a good mixture of operations is employed in the comparison. A minor criticism is that the paper does not discuss in-memory relational database technology, which might outperform MongoDB’s use of memory-mapped files. I strongly recommend this paper to those concerned with database performance.