Database Management Systems Roadmap 2026
Understand how databases work inside: relational model, SQL, storage, indexing, query execution, optimisation, transactions and recovery, following CMU 15-445.
4 stages · 7 units · Advanced · about 120 hours · Last reviewed 19 November 2026
What is a database management systems?
A Database Management System, or DBMS, is software that stores, organises and serves data while guaranteeing correctness under concurrent use and crashes. It provides a data model, a query language such as SQL, storage and indexing, a query optimiser, transactions and recovery. This roadmap studies how those internals work, following the topics of CMU 15-445.
Step-by-step database management systems roadmap
- 1
Relational Foundations
Intermediate
Model and query.
- Model
The relational model is the foundation of most databases: data is held in relations, or tables, made of typed rows and columns, linked by keys and queried declaratively. Modern SQL builds on it with joins, subqueries, common table expressions, window functions and recursion. This unit sets the vocabulary used by every later part of a DBMS.
- Model
- 2
Storage and Indexes
Advanced
How data lives on disk.
- Storage
Database storage is how a DBMS lays out data on disk and moves it through memory. Tables are split into fixed-size pages organised as heap files, with slotted pages holding variable-length tuples. A buffer pool caches pages and uses replacement policies, so the system can work with data larger than RAM.
- Indexes
Indexes are auxiliary data structures that let a database find rows without scanning the whole table. The B+ tree is the standard ordered index, supporting point and range lookups. Hash indexes give fast equality lookups, and other structures such as bitmap or skip-list indexes serve specific workloads. Each index speeds reads but adds cost to writes.
- Storage
- 3
Query Processing
Advanced
Execute and optimise.
- Execution
Query execution is the stage where a DBMS runs a chosen plan by pipelining operators such as scan, filter, join, sort and aggregate. Common processing models are the iterator or Volcano model, materialisation and vectorised batches. Operators like hash join and sort-merge join are implemented here, with attention to memory limits and parallelism.
- Optimisation
Query optimisation is the process by which a DBMS turns a declarative SQL query into an efficient execution plan. It rewrites logical plans, estimates the cost of alternatives using statistics such as cardinalities and histograms, and searches for good join orders and access paths. Poor estimates are the most common reason an optimiser picks a bad plan.
- Execution
- 4
Transactions and Distribution
Advanced
Correctness at scale.
- Transactions
Transactions group several operations into one unit that is atomic, consistent, isolated and durable. Concurrency control is the part that provides isolation when many transactions run together, using two-phase locking, timestamp ordering or multi-version concurrency control. It aims at serialisable behaviour while allowing as much parallelism as possible.
- Reliability
Database reliability covers how a DBMS survives crashes and, in distributed systems, machine failures. Write-ahead logging records changes before pages reach disk, and the ARIES algorithm uses analysis, redo and undo phases to recover. Distributed databases add replication, partitioning and commit protocols such as two-phase commit.
- Transactions
Frequently asked questions
Ready to practise for interviews?
Practise database management systems questions with an AI mock interviewer and get feedback on your answers.