Graph analytics in Aura

If your data is in Aura or you plan to move it to Aura, you have several options to run graph analytics:

If your data is in a self-managed Neo4j DBMS or a non-Neo4j data source, you can still use Graph Analytics Serverless.

Graph Analytics plugin

AuraDB Professional

The Graph Analytics plugin allows you to use the Graph Data Science library in any AuraDB Professional instance with the following requirements:

  • Neo4j version 5 or later

  • 4 GB of memory or more

  • All supported cloud providers and regions

You can enable or disable Graph Analytics when necessary, both during instance creation and on an existing instance. It can also be changed with the Aura API.

The plugin shares compute and memory resources with the AuraDB server, so you do not incur any additional costs when you enable it.

Getting started

With the Graph Analytics plugin enabled:

Graph Analytics Serverless

Graph Analytics Serverless allows you to use the Graph Data Science library regardless of where your source data is stored. It runs in Aura as a dedicated service optimised for analytics workloads, with no memory or compute resources shared with your data store.

You can enable, disable, and configure Graph Analytics Serverless on the organization level in the organization settings. The details on any running sessions are available in the Graph Analytics page.

From an AuraDB instance

AuraDB Business Critical AuraDB Virtual Dedicated Cloud

Graph Analytics Serverless is currently unavailable via any of the cloud provider marketplaces.

You can use Graph Analytics Serverless from any AuraDB Business Critical or AuraDB Virtual Dedicated Cloud instance with the following requirements:

  • Neo4j version 5 or later

  • All supported cloud providers and regions

Getting started

With Graph Analytics Serverless enabled in your organization, you only need to create Aura API credentials before you can get started.

With the Aura API credentials available:

  • If you use the Neo4j Browser or run Cypher queries through a Neo4j driver, start with the example in the Graph Data Science documentation.

  • If you use Python, start with one of the Python client tutorials depending on whether your data is in AuraDB, in a self-managed Neo4j database, or a non-Neo4j data source.

AuraDS

AuraDS Professional AuraDS Enterprise

AuraDS is the fully managed version of the Graph Data Science library where the Graph Analytics plugin is deployed by default.

In an AuraDS instance, Graph Analytics is always on. The plugin shares compute and memory resources with the AuraDS server.

AuraDS instances have the following features:

  • Upgrades and patches are automatically applied.

  • Can be seamlessly scaled up or down.

  • Can be paused to reduce costs.

Plans

AuraDS offers the AuraDS Professional and AuraDS Enterprise subscription plans. The full list of features for each plan is available on the Neo4j Pricing page.

Updates and upgrades

AuraDS updates and upgrades are handled by the platform, and as such do not require user intervention. Security patches and new versions of GDS and Neo4j are installed within short time windows during which the affected instances are unavailable.

The operations are non-destructive, so graph projections, models, and data present on an instance are not affected. No operation is applied until all the running GDS algorithms have completed.

Comparison

Graph Analytics plugin
(AuraDB Pro)		 Graph Analytics Serverless
(AuraDB BC)		 Graph Analytics Serverless
(AuraDB VDC)		 AuraDS

Maximum memory

Up to 128 GB

Up to 128 GB

Up to 512 GB

Up to 384 GB

Resources for analytics

Shared with DB (optimized for DB)

Dedicated

Dedicated

Shared with DB (optimized for analytics)

Number of concurrent sessions

-

Up to 10

Up to 100

-

Data sources

Same server

AuraDB BC; self-managed Neo4j DBMS; custom

AuraDB VDC; self-managed Neo4j DBMS; custom

Same server; custom

DBMS restart behavior

No downtime; projected graphs and trained models are not retained

No downtime; projected graphs and trained models are unaffected

No downtime; projected graphs and trained models are unaffected

Short downtime; projected graphs and trained models are restored