NetworkDisk: On-disk graph manipulation

NetworkDisk provides a way to manipulate graphs on disk. The goal is to be as compatible as possible with (Di)Graph objects of the NetworkX package, but lifting memory requirements and providing persistence of the Graph.

We aim to achieve full retro-compatibility with NetworkX core methods to ensure that most algorithms will work as-is. Some algorithms are not suited for on-disk graphs and will thus have poor performance. Some will be fast enough to be used as-is.

Audience

The target audience for NetworkDisk are users of NetworkX that want to manipulate on-disk graphs without worrying about database-specific technology or learning a new database language. No knowledge of databases is required for simple use of this module. A good knowledge of the backend is only required for advanced usage and performance tuning.

Pro and cons

There are several motivations to use on-disk graphs rather than in-memory graphs.

  1. Lack of resources: The graph is too large to fit in RAM.

  2. Persistence requirements and graph sharing: The graph must be saved on disk. You could combine the networkx and pickle modules to do that instead, but NetworkDisk saves the graph automatically, avoiding version conflict and allowing a DB user to directly access the graph DB. Also, loading a graph does not require any parsing or pre-processing.

  3. Transaction support: If an algorithm changes the graph and then fails, you wish to rollback the partial changes instead of keeping them.

  4. Concurrency control: You want many users to access the same graph.

  5. Symbolic manipulation of subgraphs: You may want to transform the Graph implicitly without actually computing the transformation.

The main drawback of using NetworkDisk rather than NetworkX is the performance loss. For direct graph manipulation, the expected penalty is of one order of magnitude (at least ten times slower) and for complex graph algorithms the penalty can be much worse.

Designing algorithms for NetworkDisk

It is possible to adapt standard graph algorithms to be used efficiently on disk. It requires a specific attention to memory management as well as avoiding repeated calls to the Backend as much as possible.

You should avoid graph algorithms which must either copy or go through the entire graph to build a dedicated DataStructure in RAM. Graph exploration algorithms (shortest path, random walk for instance) should have usable performance.

Schemas

NetworkDisk can be thought of as a dedicated ORM for NetworkX. That is, it is a mapping from the structure underlying NetworkX graphs to a relational database. The goal of this mapping is to reduce the cost penalty of performing many disk accesses.

In the codebase of NetworkDisk, we do not enforce one specific schema mapping. It can be adapted to an already existing database. It is however a complicated task to design a schema mapping that provides a correct implementation Backend. Therefore, some default fixed schemas are proposed and ready to use.

RoadMap

MultiGraph and MultiDiGraph

The NetworkX implementation provides classes for MultiDiGraph and MultiGraph. We do not provide them yet. Some specific work has to be done to adapt the current schema of Graph and DiGraph to these situations.

Other Database Engines

So far, NetworkDisk only supports SQLite local databases. A PostgreSQL version is high on the “TODO” list of the project. Other database engines are possible and could be easily adapted.

At its core, NetworkDisk was built with flexibility of the backend technology in mind. Adapting from SQLite to other backends requires some routine work. The core difficulty is to provide adjusted graph mappings that take advantage of the specificities of the engine. For instance, PostgreSQL supports the JSON datatype, which could be useful to improve performance.

Cache Policy

A crude Cache policy is provided in NetworkDisk without any control either on the memory consumption nor on any time limit to discard cached elements.

Abstract cache policies with some reasonable implementations could be added to improve this state of affairs.

Bulk cache warmup could also be of interest to start without an empty cache without having to perform many small queries.

NoSQL relational Engines

Storing graphs in a Graph Database Engine is also a possibility. If the data-model fits, it could take advantage of many graph-dedicated optimizations. Other type of DataBases could help as well (for instance distributed Key-Value stores paired with indexation engines).

A full study system per system is required and will be the subject of further works.

Indexable data fields

NetworkX does not provide any indexation properties nor any data-graph operator. Some dedicated work even on NetworkX base classes could be performed.

Graph transformation and Regular Path Queries

One of the key features of NetworkDisk (compared to standard graph databases) is the possibility of performing graph transformations through query rewriting. The performance of such a transformation depends of the inner complexity of the rewriting, the efficiency of the query optimizer, and the possibility of complex indexation approach.

A completely flexible approach to these operations is thus doomed and it should be consider separately for each backend.

Free software

NetworkDisk is free software; you can redistribute it and/or modify it under the terms of the 3-clause BSD License. We welcome contributions. Join us on GitLab.

Date:

Apr 04, 2023

Indices