Performance-Storage Trade-Off Analysis of MongoDB Indexing Strategies Using Large-Scale E-Commerce Data

Abstract

Indexing strategy in document-oriented databases represents a fundamental design decision that directly shapes both query performance and storage consumption, yet empirical evidence quantifying this trade-off at production-grade scale remains limited in existing literature. This study evaluated three indexing configurations in MongoDB, namely no index, single-field index on product identifier, and compound index on product identifier together with order status, across four collection sizes of 1 million, 5 million, 10 million, and 13 million documents derived from a synthetic large-scale e-commerce dataset obtained from Kaggle containing approximately 13 million order-line records. A controlled benchmarking procedure was employed in which each indexing condition was tested under identical query workloads repeated 30 times per configuration, with query execution time, index creation time, and index storage size recorded as evaluation metrics. Results showed that unindexed collections produced full collection scans with mean execution times scaling from 1,247.70 ms at 1 million documents to 33,300.07 ms at 13 million documents, while both indexed conditions reduced execution times to single-digit milliseconds by activating index scan paths. For multi-predicate queries, the compound index outperformed the single-field index by a factor of 6.8 at full scale, recording 1.33 ms against 9.03 ms, while incurring only 1 MB of additional storage overhead. These findings indicate that compound indexing represents the most balanced strategy for high-volume e-commerce query workloads, delivering substantial performance gains at negligible additional storage cost.