Ultipa leverages an index mechanism to accelerate the retrieval of nodes and edges by creating index tree of node/edge properties in the disk, which is comparable to traditional databases. On top of that, Ultipa index also speeds up the filtering of nodes and edges in queries such as A-B and K-Hop.

Ultipa provides a native full-text search engine to break long text in node/edge properties, and conduct efficient and fast keyword search against the revert index of these word-segmentation results that are stored in the disk. It is suitable for various queries on graph from metadata query to A-B path, K neighbor and template queries.

Ultipa's LTE method is designed to improve the filtering speed of path and neighborhood queries by loading node/edge properties to the high-performance computing engine of Ultipa, after which the I/O of disk is tremendously reduced.

Index v.s. LTE

Index is similar to the path query acceleration mechanism LTE (Load to Engine), they both improve the query performance by consuming a certain amount of disk space and ensuring the created content can be persisted, but still, they differ in some aspects.

• Diff 1：Acceleration Object

  • LTE: the main acceleration object is path query, such as A-B query, K-Hop query. Any query that involves path (depth >= 1) is good to use LTE.
  • Index: the main acceleration object is metadata query, ranging from querying for nodes and edges alone to filtering nodes and edges in paths.

• Diff 2：Implementation Principle

  • LTE: loads property index to Ultipa computing engine so that the engine can filter properties directly when querying, thus the dependency on disk-based I/O is lowered.
  • Index: creates various index trees and uses data structures in the persistent storage, therefore the query that leverages index is accelerated and does NOT cast burden to memory.

• Diff 3：Memory and Disk Usage

  • LTE-ed property: primarily consumes memory, and the occupied space is determined by the number of values of the property; and the values are kept in disk(s) persistently for the automatic reloading after the reboot of instance.
  • Index tree of index: primarily consumes disk space.

For properties that created index as well as LTE-ed:

• LTE is first used in path query
• General index is first used in metadata query