COMPLETED · ECS 165A · 5-PERSON TEAM
L-Store: A Lineage-Based Storage Engine for Unified OLTP and OLAP
A five-person academic team build of L-Store, the lineage-based storage architecture from Prof. Mohammad Sadoghi's research: columnar base and tail pages, B+ tree indexing, a buffer pool, a full query layer, and two-phase-locking concurrency control, built from scratch over a 10-week quarter.
THE STORY
Personal projects and solo work are easy; a multi-layered, 10-week academic project with four other people, each with their own lives and schedules, was not. GitHub branches, pull requests, committing changes, concurrent edits, errors, debugging, test files: the logistics were harder than the code itself, and much more rewarding to get right. It felt closer to the work done in corporate settings with teams, and to what a professional environment actually expects, far beyond the coding and conceptual execution itself.
FEATURES
Disk Persistence
Page files + metadata serialized to disk with crash recovery
Buffer Pool
LRU/MRU eviction policy with a 1000-page in-memory cache
Page Range Tables
Base + tail record architecture for efficient updates
Query Layer
Full CRUD + aggregation + record versioning
Transactions
2PL no-wait locking with rollback on conflict
Concurrency
Consistent data under heavy concurrent reads and writes
CODE
Private repo (course project) · available on request