Overview
Creating an index for a property involves generating an index tree stored on disk, which enables quicker retrieval and filtering of nodes and edges.
Ultipa's exclusive full-text index performs word segmentation for a textual property, and establishes a reverse index based on the results of the segmentation. This index, stored on disk, enables efficient keyword searches and is ideal for text-based filters.
Additionally, Ultipa facilitates the loading of a property and its index into its high-performance computing engine, known as LTE (Load To Engine). This accelerates the filtering of nodes and edges by minimizing disk I/O time.
Index vs. LTE
Both the index and LTE improve query performance by consuming disk space and persisting the created content. However, they differ in some respects.
Acceleration Object |
Index | LTE |
|---|---|---|
Node and edge queries, i.e., find().nodes() and find().edges() |
✓ | ✕ |
| The filtering of the start nodes in other queries (path, khop, etc.) | ✓ (Precedence) | ✓ |
| The filtering of nodes and edges other than the start nodes in other queries (path, khop, etc.) | ✓ | ✓ (Precedence) |
| Properties used in algorithms | ✕ | ✓ |
| Index | LTE | |
|---|---|---|
| Implementation Rationale | Creates index trees and uses data structures in persistent storage, accelerating queries without burdening memory | Loads property to Ultipa computing engine, allowing direct property filtering during queries and reducing dependency on disk based I/O |
| Memory and Disk Usage | Index tree is saved on disk | Consumes some memory depending on the type and size of the property; values are also kept on disk persistently for automatic reloading after instance reboot |
A property can be both indexed and LTE-ed to suit different scenarios. Both index and LTE will be automatically updated.