Neo4j’s desire to offer solutions to data scientists is no secret. In MagIT columns, Emil Eifrem, CEO of Neo4j, argued that the development of artificial intelligence offered significant growth prospects for his company.
In this regard, the publisher launched a data processing platform two years ago, under the name of Graph Data Science (GDS). Its fully managed distribution, accessible since April 12, 2022, is called AuraDS. It is based on the new available version 2.0 of GDS and the NoSQL Neo4j DBMS. Everything is hosted on the Google Cloud Platform (GCP).
AuraDS or Graph Data Science on demand
AuraDS encapsulates the enterprise edition of Graph Data Science. Like GDS, AuraDS includes more than 65 graphics algorithms associated with a Python client. These tools should lead to the development of machine learning and analytical use cases such as fraud detection, route optimization, product recommendation or even customer knowledge.
In its documentation, the editor specifies that the tool provides the means to prepare the data, train the models from pre-trained algorithms, and deploy them in production.
With AuraDS, Neo4j aims to simplify the work of data scientists by providing them with a drag-and-drop interface for uploading data (from a CSV file or the AuraDB import utility), modeling it into graphics. thanks to the Bloom data tool). and configure processing flows.
Like GDS 2.0, the platform has a catalog dedicated to models, and another for pipeline processing. Indeed, there are two types of algorithms and therefore two calculation methods with this tool: the first is used to process the data linked to the nodes of a graph, the second to the edges that connect the nodes.
In addition, AuraDS supports Neo4j’s Apache Spark and Kafka plugins for batch or streaming data processing.
The managed version requires, the publisher offers to automate the control of workloads, patches and backups as well as updates of this product. An MLOps feature should automate the management of model backups and restores. Data is encrypted at rest and in transit, while all backups are retained for 180 days.
Neo4j provides access to two types of instances: running and paused. A paused instance would reduce costs by 80% compared to a running instance.
The consumer pay-as-you-go business model is based on a third-party “self-initiated” billed prorated over monthly consumption, starting at $ 0.125 per GB of RAM per hour and $ 0.025 per GB of RAM per hour for paused instances . . AuraDS Enterprise is available in pre-release after signing an annual contract.
When the enterprise edition of GDS supports an “unlimited” amount of storage, RAM, and vCPU, AuraDS offers access to instances that include 8 to 96 GB of RAM, 2 to 20 vCPU, and only 16 and 192 GB of storage space.
AuraDS Enterprise will increase two of these limits to 256 GB of RAM and 40 vCPU. Please note that unlike AuraDS, this Enterprise version is not multi-tenant. It must be deployed behind a specific VPC.
A SaaS platform, but not without a server
The choice of RAM as the main unit of measurement is no coincidence: with Neo4j, data graphics reside in memory. However, customers cannot decide exactly how much RAM or computer they would need.
With AuraDS, there are seven instances of different sizes. An instance with 2 vCPUs, 8GB of RAM, and 16GB of storage costs $ 1 per hour (and 20 cents paused). The price increases by $ 1 per hour (and therefore 20 cents when instances are paused) between the first four instances. This price difference increases to $ 2 (and 40 cents at rest) per hour between the fifth and sixth instance. The best-equipped instance (20 vCPU, 96 GB of RAM, 192 GB of storage space) costs $ 12 per hour and $ 2.40 paused. Neo4J does not specify whether it is possible to stop an instance without losing workload-related backups. Also, the publisher does not guarantee an SLA. However, subscribing to AuraDS Enterprise will allow you to negotiate terms of service and get volume discounts.
Neo4j pampers its association with Google Cloud
You can pay and receive your invoice through the AuraDS console or the Google Cloud console from the Marketplace. This second option, the result of the privileged collaboration between Neo4j and GCP, allows customers of the cloud provider to subscribe to the data science service with their credits obtained by commitment of use.
Precisely, AuraDS can be integrated with Vertex AI for data science teams that have decided to manage their processing pipelines from the GCP development environment.
However, this exclusivity from AuraDS to GCP is temporary. The publisher expects to offer this SaaS product to AWS and then to Azure, as planned with AuraDB Enterprise.
Especially because Neo4j is not the first to offer a similar service: its young competitor TigerGraph had launched its offer in April 2021 at GCP.