MongoDB Data Modeling: Embedded vs Reference Selection Guide

Logical Structure

Storage Engine Comparison

Engine	Status	Use/Notes
WiredTiger	Default	Cache≈50% RAM, compression, transaction support
MMAPv1	Historical/no longer used	Only for background reference
In-Memory	Enterprise	Low latency/not persistent

WiredTiger

• MongoDB 3.2 and later default storage engine
• Document-level concurrency control, supports multi-document transactions
• Supports compression algorithms (snappy and zlib)
• Defaults to approximately 50% of physical RAM

Data Model

Embedded Model

Store associated data within the same document structure.

Advantages:

• High query performance
• Easy data consistency guarantee
• Avoid complex join operations

Applicable Scenarios:

• Data relationships are 1:1 or 1:few
• Embedded data usually needs to be queried together with parent document

Referenced Model

Establish association relationships by storing references to other documents.

Implementation:

• Manual reference: Store _id field of target document
• DBRef reference: Use MongoDB’s DBRef standard format

Applicable Scenarios:

• Data relationships are 1:many or many:many
• Associated data needs frequent independent updates

How to Choose

Choose Embedded

1. Data objects have containment relationship
2. Data that needs to be read together frequently
3. Has map-reduce, aggregation requirements

Choose Reference

1. Embedded data would cause a lot of data duplication
2. Need to express relatively complex many:many relationships
3. Large hierarchical result datasets

Minimal Runnable Example

// Insert data
db.authors.insertMany([{_id:1,name:"Li Si"},{_id:2,name:"Wang Wu"}]);
db.posts.insertMany([
  {_id:101,title:"MongoDB Guide", author_id:1, tags:["db","mongo"]},
  {_id:102,title:"Modeling Practice", author_id:2, tags:["design"]}
]);

// Create index
db.posts.createIndex({author_id:1});

// $lookup query
db.posts.aggregate([
  {$match:{tags:"db"}},
  {$lookup:{from:"authors", localField:"author_id", foreignField:"_id", as:"author"}}
]);

explain() Reading Quick Reference

Field	What to Look For	Action
stage	COLLSCAN vs IXSCAN	COLLSCAN appears → add index
nReturned / totalDocsExamined	Ratio	Ratio≪1 → rewrite query/composite index
indexName	Which index hit	Adjust order/composite fields