dbms

Language: C++11

Built my own database management system from scratch, with the help of a teammate. This project was in the scope of the class “Principles of Data Management”, at UC Irvine, and was divided into the following parts:

  1. Disk I/O Page-Based Manager
  2. Record-Based File Manager
  3. Relation Manager
  4. Index Manager
  5. Query Engine

1. The Disk I/O Page-Based Manager

  • Provides higher-layered managers with tools to perform I/O operations in terms of pages (eg: open/close file, create/destroy file and read/write pages to the file).

2. Record-Based File Manager

  • Responsible for inserting, deleting and updating records within a given page-based file. Records are identified and located within a file by a pair (pageNumber, recordSlot) called RID.
  • The attribute types supported are: INT, REAL and VARHCAR(N). Thus, variable-length records are possible.
  • It has the responsibility of managing the free space offset within a page and guarantee $O(1)$ time for accessing the $i^{th}$ attribute of the record. For this reason, both pages and records are encoded in disk with a specific format chosen by the manager itself.

3. Relation Manager

  • Responsible for managing the database tables (creating/deleting tables and inserting/removing tuples)
  • It keeps database schema in the system catalog, which is itself a table

4. Index Manager

  • Implemented using a B+ Tree that supports all the necessary operations: insertion, removal, key lookup and range scan.
  • Every node in the tree, intermediate or leaf node, corresponds to a page of size 4096 bytes and it must be at least half full – requirement for B+ trees to reduce sparsity and fragmentation.
  • For simplicity, we don’t care about keeping the above property whenever deleting a record. This is because merging nodes is a complex task (splitting in insert are simpler) and deletes are not frequent.

5. Query Engine

  • Provides the functionality to answer SQL queries
  • The following relational operators were implemented:
    • Filter
    • Projection
    • Aggregate (with “Group by”)
    • Join, using:
    • Block-Nested Loop Join
    • Index-Nested Loop Join

Related