A fixed storage distributed graph database hybrid with at-scale OLAP expression and I/O support of a relational DB: Kinetica-Graph

TL;DR

This paper presents Kinetica-Graph, a distributed graph database with fixed storage, hybrid OLAP capabilities, and efficient I/O support, featuring a unique topology and non-blocking architecture for scalable geo-spatial and social network analysis.

Contribution

It introduces a novel fixed storage topology, non-blocking client-server communication, and a balanced partitioning scheme with a distributed shortest path solver.

Findings

Supports hundreds of PostGIS functions.

Achieves minimal communication overhead.

Demonstrates efficient re-balancing and shortest path computation.

Abstract

A distributed graph database architecture that co-exists with the distributed relational DB for I/O and at-scale OLAP expression support with hundreds of PostGIS compatible geometry functions will be discussed in this article. The uniqueness of this implementation stems mainly from its double link topology structure for its fixed storage characteristics independent from the variance in node-to-edge connections. Another note-worthy contribution of this implementation is its non-blocking client-server communication architecture among its distributed graph servers. A non-bottlenecking partitioning scheme based on duplication of nodes is also implemented ensuring minimal communications using distributed filtering on geo-spatial, random and explicit sharding choices. Finally, an efficient re-balancing algorithm followed by a distributed shortest path solver will be demonstrated with examples from both geo-spatial and social networks.