TEGRA: Efficient Ad-Hoc Analytics on Evolving Graphs
summary
TEGRA exposes evolving graphs via a Timelapse API and stores versions with a persistent adaptive radix tree, enabling snapshot reads without replaying deltas. It implements diff/expand/merge incremental execution so only changed vertices recompute, falling back to full runs when changes are large. Prototype runs on Apache Spark and, versus Differential Dataflow, GraphOne, Aspen and Chronos, shows two orders of magnitude faster random snapshot retrieval and 8–30× speedups on ad‑hoc windowed graph analytics.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:30Meet Carol
- 0:54How Carol does this
- 2:29Challenges
- 3:49Timelapse Abstraction
- 4:39Snapshot Storage
- 5:18Optimizing Storage
- 6:07Distributed Graph Snapshot Index DGSI
- 8:59Implementation & Evaluation
- 9:31Ad-hoc Retrieval
- 10:15Ad-hoc Analytics
- 10:51Computation Model Benchmarks
- 11:38Summary
related talks
27mGraphOne: A Data Store for Real-time Analytics on Evolving Graphs
35mBuilding Real-Time Graph Analytics Platforms
30mAerospike Graph: Massively Scalable, High Performance
29mTigerGraph, Connecting the World
38mA Tale of Two Graph Engines on Spark
49mModern Graph Databases
23mScaph: Scalable GPU-Accelerated Graph Processing with Value-Driven Differential...
20mBridging the Gap between Relational OLTP and Graph-based OLAP