The #1 Platform for Connected Data

WHITE PAPER

The Graph Technology

Buyer’s Guide
What You Should Know before
Selecting a Graph Technology Solution
Amit Chaudhry, Vice President, Neo4j

neo4j.com
 The #1 Platform for Connected Data

White Paper

The Graph Technology

TABLE OF CONTENTS

What Are Graph Databases

Buyer’s Guide
Good for? 4

Traditional Technology Choices

Do Not Consider How Data Is
Interrelated 5
What You Should Know before Selecting
Collections vs. Connections 6
a Graph Technology Solution
Property Graphs Are
Intentionally Simple 7 Amit Chaudhry, Vice President, Neo4j
Benefits of Graph Databases 8

Structure 10
Introduction
Open Source Foundation & The graph technology market is heating up with a wide variety of database and analytics
Community 10
vendors staking claim to supporting graph (connected data) capabilities.

Native Graph Storage 11 The challenge for the modern technology buyer is weeding through mountains of material
in order to make an informed buying and implementation decision. This challenge
ACID Compliance 11 is compounded because graph technology is a new way of managing data. Data
consistency, performance, scalability and other important characteristics of any database
Graph Query Languages 13 technology are all much different with graph databases than with traditional data
management platforms like relational databases (RDBMS).
Hybrid Transactional-Analytic
Platforms (HTAP) 15
This buyer’s guide is designed to help expedite those decisions and explain what makes
purchasing this type of technology so different from other database purchase decisions.
OLTP Applications 15
For the full list of buying criteria and questions, skip to Part 2 on page XX. For readers who
aren’t yet familiar with the basics of graph technology, we’ll start with a brief introduction.
Query Traversal Performance
Benchmark Comparisons 17

Graph Analytics 18

Data Integration & Ingestion 18

Graph Platform with Tools

& Support for All Types of
Users 19

Developer Tooling 20

Visualization Capabilities 21

Deployment, Scaling &

Administration 22

Business Model, Focus

& Staying Power 26

Conclusion 28

1 neo4j.com
The Graph Technology Buyer’s Guide

Part 1:
What Are Graph
Databases Good for?

2 neo4j.com
The Graph Technology Buyer’s Guide

Viewed through a What Are Graph Databases Good for?

technological lens, Graph database technology is specifically designed and optimized for highly
graph databases interconnected datasets to identify patterns and hidden connections.
tackle the most Graph data stores are intuitive because they mirror the way the human brain thinks and
harrowing of data maps associations via neurons (nodes) and synapses (relationships). A graph database
problems – ones efficiently stores and queries connected data in a node-and-relationships format. As
a result, graph technology excels at problems where there is no a priori knowledge of
that often linger at path length or shape by using graph storage and infrastructure to find neighboring data
the root of project efficiently.
failures and delays. The most common graph use cases and solutions include:

• Fraud Detection & Analytics: Real-time analysis of data relationships is essential

to uncovering fraud rings and other sophisticated scams.

• Artificial Intelligence & Machine Learning: Artificial intelligence (AI) winners

and losers will be decided based on who harnesses context within data for a true
competitive advantage.

• Real-Time Recommendation Engines: Graph-powered recommendation

engines help companies personalize products, content and services by building a
contextualized map of offers using both historical and real-time data.

• Knowledge Graphs: Graph-based search tools tap into your organization’s

institutional memory. They are also used for better digital asset management.
Moreover, knowledge graphs are the basis for many natural language processing
(NLP) and AI solutions.

• Network & Database Infrastructure Monitoring: Graph databases

are inherently more suitable than RDBMS for making sense of complex
interdependencies central to managing networks, data centers, cybersecurity and
IT infrastructure.

• Master Data Management (MDM): The schema-optional graph database model

allows you to organize and manage your master data with flexibility. It also lets
you harness real-time insights and a 360° view of your customers, products and
employees.

What these use cases have in common is that their success requires solving complex
problems with dynamic and interconnected datasets. To this end, we should reframe
the question, “What are graph databases good for?” as a technical one.

Viewed through a technological lens, graph databases tackle the most harrowing of
data problems – ones that often linger at the root of project failures and delays. These
include:

• Vastly different views of the data model between business and technology teams,
which result in misunderstanding and miscommunication.

• Lack of schema flexibility and adaptability, making it hard to respond to changing

business requirements both during a project and after a system has gone live.

• The “JOIN problem” which occurs when queries become so tangled that even
powerful databases with massive amounts of hardware resources grind to a halt in
their attempts to bring the resulting data together.

With the right choice of technologies, graph databases introduce a new way of looking
at data by promising to significantly address all of these issues. They seek to take you
beyond handling sheer volumes of relatively simple data and towards revealing the
interconnected complexities latent in your data – and deriving bottom-line value from
them.

3 neo4j.com
The Graph Technology Buyer’s Guide

Well-established Traditional Technology Choices Do Not

entities such as
Consider How Data Is Interrelated
NASA, eBay, UBS and
many of their peers Today’s data and applications require elasticity, agility, speed and interconnectivity.
use graph technology Despite the name, relational databases are not well-suited for modeling and storing
today’s highly connected and agile datasets. RDBMS demand slow and expensive
to improve schema redesigns that hurt agile software development processes and hinder your
their customer ability to scale and innovate quickly.
experiences and Traditional RDBMS technology has a difficult time expressing and revealing how real and
increase their virtual entities are related. Columns and rows aren’t how data exists in the real world.
Rather, data exists as rich objects and the relationships between those different objects.
competitiveness.
The theory of connected networks (graph theory) has been a mathematical discipline for
nearly three centuries, but few technologies have harnessed these theoretical models
for the purpose of data storage and analysis.

Early on, a handful of companies have truly tapped into the power of graph technology
as the driver of their businesses, including Google, Facebook, LinkedIn and Microsoft as
well as upstarts like Airbnb and Uber. Well-established entities such as NASA, eBay, UBS
and many of their peers use graph technology to improve their customer experiences
and increase their competitiveness.

Today, graph-powered applications are used by more than 75% of the Fortune 500,
including:

• 7 of the world’s top 10 retailers

• 3 of the top 5 aircraft manufacturers

• 8 of the top 10 insurance companies

• All of North America’s top 20 banks

• 8 of the top 10 automakers

• 3 of the world’s top 5 hotels

• 7 of the top 10 telcos

These successes serve as a strong indicator of graph technology’s impact on both

innovation and the bottom line.

4 neo4j.com
The Graph Technology Buyer’s Guide

By eschewing data Collections vs. Connections

relationships and
providing simple SQL & NoSQL Systems Focus on Data Aggregation &
programmatic APIs, Collection
NoSQL systems make Collection-centric storage designs as implemented by SQL and Not only SQL (NoSQL)
it easy for developers databases are designed to efficiently divide and store data.

and administrators In SQL’s case, the normalization of data into a tabular schema aims to minimize storage
of duplicate data objects, types and values. These systems were born during the era of
to work with simple
scarce physical memory and expensive disk-based storage, designed to avoid managing
data in a way that can often-redundant data objects such as physical location addresses for shipping, billing,
easily scale. homes, offices, destinations, businesses, etc. For example, all of these data objects
included common, redundant data such as a country and its provinces or states, or
telephone area codes and postal codes.

The original relational databases were designed to minimize storage of duplicative

data values because disk space was costly. (Ironically, each normalization incurs a cost
in relationship storage in the form of JOIN tables, which native graph databases have
managed to eliminate through the use of pointers.) The RDBMS design achieved this
consolidated, normalized goal by linking tables of data via foreign keys to their associated
records from other tables. This is why a relational dataset modeled into a graph often
shrinks by an order of magnitude or more, maintaining the full richness of the data
without redundant data storage.

NoSQL systems like document, wide column and key-value data stores carry those
concepts forward (and backward) by simplifying their models in exchange for higher
levels of scale and simplicity. By eschewing data relationships and providing simple
programmatic APIs, NoSQL systems make it easy for developers and administrators to
work with simple data in a way that can easily scale.

A lack of concern about relationships leads to looser data guarantees, plain APIs and
straightforward scaling schemes. Data is easily spread out and just as easily retrieved,
without the need to maintain the integrity of related data that’s written across a
distributed storage or a cluster of machines and without needing to concern itself with
the performance of distributed JOINs across those machines.

These systems take on the “store and retrieve” problem at scale for simple data, and
their architectures reflect this, as does the set of problems they are equipped to address.
However, none of these systems focus on interrelated, contextualized data or how that
data might be traversed to reveal unobvious relationships, as explained below.

5 neo4j.com
The Graph Technology Buyer’s Guide

A property graph Graph Systems Focus on Data Connections

is a data model By contrast, graph database technologies focus on how data elements are interrelated
designed to express and contextualized as connected data.
data connectedness Connected data is the materialization and harnessing of relationships between data
as nodes connected elements, which is modeled as a property graph.
via relationships to A property graph is a data model designed to express data connectedness as nodes
other nodes, where connected via relationships to other nodes, where both nodes and relationships can have
properties attached to them (which in turn can be indexed). For example, devices on an
both nodes and enterprise network might be modeled as nodes, with properties for their attributes (such
relationships can as throughput) and relationships pointing to other adjacent devices.
have properties In the graph model, data relationships are persisted so they can be navigated or
attached to them traversed along connected paths to gain context. Relationships are both typed and
(which in turn can be directional.

indexed). The context provided by these data connections is essential to identifying friendships,
making relevant real-time recommendations, attaching adjacent ideas and detecting fraud
by following money trails. Without relationships as first-class data entities, all of these use
cases become extremely difficult to execute.

Property Graphs Are Intentionally Simple

• You can draw property graphs on whiteboards and map that design directly into a
graph database.

• You can change or update a property graph easily, because its agile design
eliminates most of the structural overhead of traditional database schemas.

• You can quickly program property graphs because their query language expresses
and follows relationships.

• You can visualize and navigate property graphs efficiently by following the
relationships on their paths to context.

• You can rapidly determine data context when property graph queries are
executed in hyper-fast native graph platforms built on reliable, scalable database
architectures.

6 neo4j.com
The Graph Technology Buyer’s Guide

Connected data in Benefits of Graph Databases

property graphs
enables you to • Simple and natural data modeling: Graph databases provide flexibility for data
modeling, depending on relationship types. Since the graph model comes with
illustrate and no inherent rules, graph data stores add as much or as little semantic meaning
traverse many as the domain requires. This occurs without any constraints like normalization or
relationships and restructuring of the data using denormalization.

find context for your • Flexibility for evolving data structures: Graph technology provides flexible
next breakthrough schema evolution. In a constantly changing data environment, you need the option
to add or drop data entities or relationships as well as extend or modify your data
application or model. Graph databases allow for evolving data structures that match today’s agile
analysis. development environments.

• Simultaneous support for real-time updates and queries: A graph data store
and its model allow real-time updates on graph data while supporting queries
concurrently.

• Better, faster and more powerful querying and analytics: Graph data stores
provide superior query performance with connected data using native storage and
native indexed data structure.

Connected data in property graphs enables you to illustrate and traverse many
relationships and find context for your next breakthrough application or analysis.

7 neo4j.com
The Graph Technology Buyer’s Guide

Part 2:
The Buying Criteria for
Graph Technology

8 neo4j.com
The Graph Technology Buyer’s Guide

Open source Structure

technology helps
drive adoption, The structure of this document allows the reader to understand the buying criteria to
consider, given their desired use case. Each vendor you consider should be able to
ensure quality and robustly answer your questions according to these criteria.
assist market fit as
Graph DBMS buying criteria is often mischaracterized as being akin to purchasing any
the software matures traditional database. To some degree that is true, with shared questions such as:
version by version
rr Does it store data safely and redundantly?
with the feedback
rr Does it have security features?
– and sometimes
contribution – of a rr Does it have an easy-to-use query language?
global community. rr Does it scale?
rr Is it embeddable?
rr Does it run in the cloud?
rr Does it work with my existing applications?
All of these questions are indeed valid. There are, however, important requirements
related to supporting and persisting data relationships that make graph database
management systems and platforms unique, therefore requiring specific buying criteria.
These criteria are explained below.

Open Source Foundation & Community

The most successful infrastructure software contains a strong foundation of open source
code that provides power and transparency, often catalyzed by a community of users
around that software.

Open source technology helps drive adoption, ensure quality and assist market fit as the
software matures version by version with the feedback – and sometimes contribution – of
a global community. Beyond Oracle, Microsoft SQL Server and a handful of other early
RDBMS, very few proprietary database technologies survive when compared to their open
source counterparts like Postgres, MySQL, MariaDB, MongoDB and many others.

The premise of using open source as a means to validate and move a market forward is
no different in the case of graph databases. Therefore, buyers should consider the size
and enthusiasm of a graph technology’s open source community as a proxy for multiple
items, including the market impact of that technology, the likely availability of skill sets
in the market, and the flexibility of the software to evolve beyond the capabilities and
personnel of a single software vendor.

What to Look for: Open Source Community

There are a number of graph technologies that are born from open source projects. The
most notable is Neo4j, which was the first property graph database released as open
source in 2010. After nearly a decade, Neo4j’s community is thriving with millions of
downloads and hundreds of thousands of deployments.

Apache TinkerPop is another open source graph project. The TinkerPop graph traversal
engine’s v1 release was in 2011, and it became an Apache-incubated project in 2015.
Given its ease of integration, flexible storage options and permissive-use license, it has
become a common choice for NoSQL vendors when adding graph interfaces to their
products.

9 neo4j.com
The Graph Technology Buyer’s Guide

Long considered There are a number of other proprietary graph vendors as well. However, their ability to
establish market share and adoption is severely limited when compared to the Neo4j and
the “gold standard” TinkerPop open source projects.
for database
Independent sites like DB-Engines prove out this market reality. As of the publication of
transactional this guide, Neo4j’s score accounts for roughly half of the total graph market popularity
reliability, ACID score. Meanwhile, TinkerPop-based products make up approximately 40 percent of that
(Atomicity, score. The remaining 10 percent are divided among over 20 additional vendors.

Consistency, Isolation
and Durability) Native Graph Storage
compliance ensures
that a DBMS safely Native graph storage ensures that relationship information – the entity that connects one
data node to another – is persisted as a primary data element.
and reliably stores
transactional data Without native graph storage, relationship information may be lost, misconnected or
abandoned, all of which are unique symptoms of graph data corruption. Many non-
as it enters and is native graph solutions allow for (and sometimes create) graph data corruption, which is
updated within the unacceptable for mission-critical enterprise applications.
system – especially as
the data and system What to Look for: Transposition vs. Persisted Relationships
grow. When considering non-native graph technology, a buyer should ask: Is there a graph-to-
document or graph-to-column transposition step buried in the software?

Document databases often store relationships as special document collections. Columnar

databases may store relationships as column families. Neither is native, thereby increasing
complexity and slowing processing times, especially if a graph traversal query requires
multiple hops where each hop needs to be individually translated from the underlying
system.

What to Look for: Index-Free Adjacency

Native graph storage with index-free adjacency supports high-performance graph
traversals across deep graph datasets.

Index-free adjacency avoids expensive, low-level transpositions that are supported by non-
native graph storage and retrieval architectures. Index-free adjacency creates fixed-size
pointers to and from any node to another; native graph technology is therefore both fast
and predictable in its performance costs.

ACID Compliance
Long considered the “gold standard” for database transactional reliability, ACID (Atomicity,
Consistency, Isolation and Durability) compliance ensures that a DBMS safely and reliably
stores transactional data as it enters and is updated within the system – especially as the
data and system grow.

While some (NoSQL) databases eschew ACID and can get away with doing so for certain
use cases, the impact of non-ACID compliance with graph datasets is grave. Because
graphs are inherently connected, proven risks arise with non-ACID transactions, where
parts of a graph transaction successfully commit while other parts fail, leaving data in a
corrupted state.

For example, corrupted graph data can result in dangling relationships that point
nowhere, phantom properties, and nodes that can only be reached from one direction.
The ripple effect of these situations is quite staggering when you consider what happens

10 neo4j.com
The Graph Technology Buyer’s Guide

Unlike the ISO/ANSI when subsequent reads against that corrupt data influence new updates or writes, which
further spread corruption because they are based on the original corrupt, untrustable
SQL that has served values. Ultimately, significant portions of the database become unusable.
RDBMS users as
For non-native graph databases built on top of other data models, ACID compliance can
the standard query be equally complex and concerning. For example, key-value stores are optimized for high
language for 40 throughput and low latency on single-key lookups (reads) and writes. They support ACID
years, there is not yet semantics only at a single-key level. As a result, a graph data store that needs transactional
and data consistency capabilities cannot be implemented directly over native key-value
an industry-standard stores efficiently.
query language for
graphs. What to Look for: Durability
Database systems should be evaluated and rigorously tested to ensure that they lose no
data when introducing a systemic hardware failure, such as a “kill -9” process interruption,
power failure or kernel panic.

It has been observed that some native graph and multi-model database systems are
unable to recover gracefully here, both by losing transactions that the system had
previously accepted and/or failing consistency checks after recovering from the failure
itself.

What to Look for: Consistency Checking for Graphs

When you’re evaluating vendors, here are the questions you should consider when it
comes to data consistency and corruption:

rr Does the database offer any level of consistency checking for graphs?
rr What safety measures are available to ensure writes are written correctly?
rr Do these measures work as expected in a clustered, High Availability (HA)
environment?

rr Does the database vendor’s documentation contain warnings, such as beware of

ghost vertices or floating or untethered edges?

All of these questions should serve as warnings of data corruption.

11 neo4j.com
The Graph Technology Buyer’s Guide

Today, during the Graph Query Languages

emergence of
NoSQL and graph Unlike the ISO/ANSI SQL that has served RDBMS users as the standard query language for
40 years, there is not yet an industry-standard query language for graphs.
databases, it is not as
easy to find trained Consequently, graph query language support varies widely across products. A number of
graph-specific query languages have therefore been developed. Some of these support
developers who only the vendor’s own product, whereas others – Cypher and Gremlin – support a number
know each vendor’s of DBMS implementations as well as third-party graph visualization and integration tools.
proprietary query A subclass of the single-product languages use their own SQL-like dialects adapted by the
language. vendor in an appeal to users’ existing experience and comfort with SQL. Such approaches
– while appearing familiar on the surface – tend to elude the desired benefits.

These SQL-like dialects are often supported by only one database vendor, lack a
community of skilled users together with a third-party tooling ecosystem, and also tend
to miss out on the technical benefits of query languages designed from the ground up for
the property graph model.

Below is a review of more detailed considerations, but here is a summary of where most
users end up: nearly all graph database vendors offer either direct Cypher support, direct
Gremlin support or both. All graph databases that support Gremlin can now also run
Cypher, thanks to openCypher’s “Cypher for Gremlin” project, which has been donated to
the Apache Foundation under the official TinkerPop umbrella, making Cypher support part
and parcel of TinkerPop.

Also worth noting is that the vast majority of graph database users (which encompasses
developers as well as projects) are using Cypher. There is an active and concerted effort
inside of the ISO and the openCypher community to create a new standard language
under the International Standards Organization umbrella as a sibling language to SQL.
This language is inspired by and would be highly compatible with Cypher and its own
sibling graph query language derivatives.

What to Look for: Declarative vs. Imperative Languages

Declarative languages such as SQL – as opposed to imperative languages like Java – tend
to be advantageous as database query languages for a variety of reasons.

Declarative query languages are more easily learned, as the query author needs only to
instruct the database about what to retrieve and not be concerned about the low-level
details of how the database should go about obtaining it.

Declarative languages are also easier to write, read and debug for end users than
imperative languages. This declarative approach has been a key factor in SQL’s popularity
over the years. Today, the most popular, fully declarative graph query language is Cypher.

Imperative languages offer fine-grained control over every behavior of the query by
instructing the system where to go as it touches every node, almost like when one learns
hopscotch. The issue with imperative languages is that the developer must be intimately
familiar with the graph schema they are querying in order to tell the system how to
execute each traversal, which increases both the learning curve for the language as well as
the complexity of the code being written.

The most popular imperative graph query language is Gremlin, though Gremlin also
combines some declarative attributes.

Declarative languages require far more work on the part of the database vendor to build
and optimize, as they effectively render what a user is asking for into a set of low-level
imperative commands that the database carries out. TinkerPop is thus relatively easy
to port to any database backend. On the other hand, supporting Cypher requires a

12 neo4j.com
The Graph Technology Buyer’s Guide

The most successful sophisticated combination of parsing, planning and runtime work, as well as database
statistics and cost-based optimization techniques, all of which require substantial effort
graph applications to write and maintain.
help organizations
innovate more What to Look for: Compiled vs. Interpreted Instructions
quickly, generate Some graph query languages must be compiled into binary packages prior to execution
and optimize within the database. While this technique may be advantageous for optimizing the
performance of a single query, it undoubtedly increases debugging complexity and
revenue streams,
extends the time it takes to write, build and execute queries.
and improve the
Compiled queries are also a major obstacle when it comes to ad-hoc querying by end
customer experience users and end-user tools. This effectively eliminates the ability to change a query on the
by preventing fly – perhaps to just correct a typo – or to modify an ad-hoc query parameter.
unexpected
maintenance. What to Look for: Graph Query Language Standardization
The relational database market of the 1980s grew much more rapidly after multiple
vendors adopted ANSI SQL as their standard query language. This made it easier to find
and hire skilled resources for nearly any project and created more application portability
across platforms.

Today, during the emergence of NoSQL and graph databases, it is not as easy to find
trained developers who know each vendor’s proprietary query language. This will remain
true until another standard language takes shape.

Developing a new standard not only requires an abundant user community, but also
cooperation among vendors, large and small. This cooperation has not happened in the
NoSQL space, as the SQL Standards oversight team chose to simply adapt the ANSI SQL
standard to accommodate new NoSQL-style syntaxes and data types.

Graph Query Language (GQL) standards development is underway. This new standard
is emerging both from the usage and adoption preferences of the graph tech user
community, as well as the adoption and implementation of open source language toolkits
within the vendor community.

What to Look for: GQL vs. Gremlin

Apache TinkerPop enjoys popularity across a variety of vendor implementations. Gremlin
is TinkerPop’s query language and is supported by many vendors including Neo4j, who
co-authored Gremlin’s original TinkerPop engine. Gremlin’s drawbacks, however, are
explained above.

Contrast this to the openCypher project with growing support from SAP HANA, Databricks
and the Apache Spark project, Redis, Memgraph, Bitnine and Cambridge Semantics.
There is also a swell of support for GQL, as the ISO W3C (the SQL Standards body)
held a discussion on endorsing both a SQL-compatible, yet independent, Graph Query
Language based on inputs from the Linked Data Benchmark Council (LDBC), Neo4j’s
Cypher and Oracle’s Property Graph Query Language (PGQL).

13 neo4j.com
The Graph Technology Buyer’s Guide

As the software Hybrid Transactional-Analytic Platforms

industry has evolved,
(HTAP)
most IT organizations
have become wise Database use cases are generally divided into two top-level categories – those that
to the configuration support online transaction processing (OLTP) applications and those that support analytic
activities (OLAP), including data warehouses, business intelligence, data discovery, data
and execution mining, etc.
games played among
While graph database use cases can fall into these general buckets of OLTP and OLAP,
vendors, giving they often layer or blend analytic and transactional information in an approach known as
rise to the term Hybrid Transactional-Analytic Platforms (HTAP).
“benchmarketing.” For example, a graph-powered metadata management application can act as both
a knowledge graph for data lineage as well as a recommendation application. Or, a
data scientist can develop AI instructions using graph analytic algorithms and then
deploy those algorithms to a transactional AI application that makes personalized
recommendations to each user.

OLTP Applications
Common OLTP use cases for graphs include identity and access management, real-
time recommendations, social network publishing, fraud detection, network systems
operations, cybersecurity, Internet of Things (IoT), artificial intelligence, supply chain,
institutional knowledge recall, customer 360, digital transformation and metadata
management.

Often, these are combined or layered to feed one another. Moreover, these transactional
applications must integrate a variety of technologies beyond the underlying data
management system. Therefore, the graph technology must support integrations across
use cases, data interfaces, deployment environments, programming languages, user
experience and more.

What to Look for: Verified Use Cases vs. Website Messaging

Ask each supplier for example implementations and references for each of your use
cases.

If any vendor is unable to supply solid customer references, then there’s a high likelihood
that the implementations only exist in theory. The greater number of example successes
for your use case, the higher the probability of your own success in implementing a graph
solution.

Be sure to ask any vendor that provides use case examples whether the referenced
company has a separate commercial relationship with the graph provider – for example,
as an investor or business partner.

The most successful graph applications help organizations innovate more quickly,
generate and optimize revenue streams, and improve the customer experience by
preventing unexpected maintenance. Make sure your graph technology vendor knows
how and why their customers and users chose graphs for their application. Beware of
vendor inexperience and “foilware.”

Verified community and customer experience should provide higher degrees of

assurance in your search for the right vendor.

14 neo4j.com
The Graph Technology Buyer’s Guide

When data is What to Look for: Geospatial Support

imported into a Transactional graph applications are frequently designed to describe the interaction
graph, the data’s activities among people with a variety of digital and analog objects in the context of their
location, time, space and the digital networks within which they operate.
relationships must
be materialized. If these applications, filled with graphs of people, objects, locations, events and networks
sound familiar, then the system within which you store and query this information should
Furthermore, due to
support advanced data types, such as geospatial and temporal types.
the schema-optional
nature of graph What to Look for: Temporal Support
systems, data is easily Many graphs have temporal requirements since they contain information about when
changed, added and activities occurred, and therefore how they are sequenced in the graph.
removed. rr Does the graph system support time formats and calculations?

What to Look for: Full-Text Search & Indexing

Many graph applications parse and search documents and large parcels of text such
as maintenance orders, parts assemblies, research documents and even web pages. In
these cases, full-text search and indexing are critical for optimal performance.

rr Does the graph system support mapping, indexing and searching for specific text
within the graph? How is that performed?

15 neo4j.com
The Graph Technology Buyer’s Guide

A graph platform Query Traversal Performance Benchmark

should include
Comparisons
features for all
of these users, Benchmarks are a popular product comparison vehicle, but you should observe graph
offering extensions, benchmark results with a skeptical eye. As the software industry has evolved, most IT
organizations have become wise to the configuration and execution games played among
apps, libraries, vendors, giving rise to the term “benchmarketing.”
APIs and SDKs, as
When evaluating vendor-sponsored benchmarks, look for verifiable and reproducible
well as integration tests that include data, documentation, configuration and execution methods as well as
with a variety of results. Look for a standard toolkit and dataset – for graph applications this is the Linked
Data Benchmark Council’s (LDBC) array of tests.
complementary
technologies. Look for independent third parties that have performed comparisons – even when
working with the graph vendor. These comparisons are preferred over highly biased,
vendor-developed and staged tests.

What to Look for: Query Traversal Performance Tricks Played

by Vendors
When creating performance benchmarks, vendors often use tricks to improve their
product performance while simultaneously causing their competitors to perform poorly.
Here are a set of key questions to ask while evaluating any vendor benchmark:

rr Does performance deteriorate exponentially when the number of traversal hops

across the graph increases? (This is very likely due to the transposition issues
mentioned above, or whether the queries are run as executables or within a
runtime interpreter.)

rr Does the benchmark compare a pre-compiled binary query against an interpreted

query? Is that fair, or should both systems run pre-compiled queries?

rr Does the vendor fully exercise the system or simply highlight its most favorable
tasks and results? For example, are workloads mixed as both transactional writes
and analytic reads, or is only one type of activity executed?

rr Are caching strategies equal? Are both graphs fully loaded into memory (where they
will be extra fast), or is only one product fully loaded into memory while the other
swaps between cache and disk?

rr Are one-time data loading operations overly emphasized, versus multi-user or

multi-session tests, which better reflect regular usage?

rr Are benchmark sessions multi-user, or are they single sessions that utilize all
available CPU resources to parallelize query execution? How realistic is it that one
user will use the database at a time?

rr More broadly, do the benchmarks consider real-world, mixed read-write workloads?

rr Are the systems configured optimally and fairly: system sizing, memory, tuning
config parameters?

Look out for benchmarks that claim fairness by using database defaults, which are often
extremely minimal in order to support small-scale functional evaluation rather than
performance at scale.

16 neo4j.com
The Graph Technology Buyer’s Guide

A visual development Graph Analytics

environment should
be included with the Graph analytics are unlike traditional aggregation-based analysis.

database to help Graph analytics tasks search for pathways, clusters, similarities and contextualization.
developers write Given the newness of graph analytic exercises, you should place a higher degree of
confidence in vendors who offer a variety of educational, implementation and integration
clean, clear queries materials to assist data scientists and other data analysts.
and also understand
what those queries What to Look For: Publications & Educational Material
return. rr Has the vendor not only published content about graph systems, but also
published information about how to perform graph analytics exercises using graph
algorithms?

rr If not, then how do they teach users how to use any graph algorithm libraries they
might offer?

What to Look for: Robust Library of Graph Algorithms

Graph algorithms help users readily identify patterns, paths, clusters and similarities
among graph data. These algorithms are used to support AI applications and reveal
interesting patterns during analysis. Vendors should offer support for popular graph
algorithms, including documentation explaining their appropriate use and application.

Data Integration & Ingestion

Data ingestion and integration are just as important for graphs as they are for SQL and
NoSQL systems.

When data is imported into a graph, the data’s relationships must be materialized.
Furthermore, due to the schema-optional nature of graph systems, data is easily
changed, added and removed. Vendors will outline their ability to ingest high volumes of
data rapidly, which is primarily useful when a dataset first enters the graph. Enterprises
who have invested in other big data technologies will need integration to their Hadoop
installations, ideally using already-deployed data processing technologies such as Apache
Spark.

It also may become necessary for data to be streamed or imported into the graph system
on a regular basis. Therefore, the inclusion or integration with data processing tools such
as Kettle or Kafka (open source) or commercial tools from Informatica, iWay, Trifacta or
others is desirable from any graph technology vendor.

What to Look for: Data Lake Integrations

Data lakes have become commonplace across enterprises. This mega-warehouse
strategy is designed to centralize data storage in order to feed traditional data
warehouses, support a new generation of transactional and analytic applications and
ensure consistency of how data is managed, maintained and used.

Organizations who have invested in data lakes – and especially those who use Apache
Spark as their analytic processing engine – should consider how easily the graph system
integrates with that source data.

17 neo4j.com
The Graph Technology Buyer’s Guide

A well-built graph Some questions to consider for data lake integrations with graph technology:

visualization tool rr Can the graph system operate within the Apache Spark ecosystem?
allows you to explore rr Can graphs be materialized from Spark Data Frames?
the graph in a
rr Can graph structures be directly imported into the graph database system?
variety of ways, such
as filtering nodes What to Look for: Data Integration & Ingestion Technologies
and relationships, Data integration and data streaming capabilities are very desirable add-ons for graph
lassoing parts of the systems. Without them, the DBA is left with simple comma-separated value (CSV) import
graph, reviewing and functions, which could be sub-optimal.

editing properties, Look for libraries, SDKs and data integration adapters to popular data management tools.
Open source tools like Talend and Kettle should be available for a given graph system,
highlighting paths,
while commercial product support from Informatica, Ab Initio, iWay, Trifacta or Tamr is
coloring nodes desirable for organizations that have made investments in these tools.
categories and
What to Look for: CSV Import & High-Speed Ingestion
including graphic
At a minimum, most products import CSV text files. Any given graph technology vendor
icons.
should also support additional functions such as high-speed and bulk ingestion.

More advanced data population strategies include the ability to defer index building and
consistency checking until after all the data is ingested. However, development teams
should be careful not to over-consider the importance of high-speed ingestion. For most
use cases, high-volume ingestion is usually more of a setup exercise when compared to
ongoing incremental updates, which tend to be much smaller than initial data loads.

Graph Platform with Tools & Support for All

Types of Users
Most use cases require more than just a database. Therefore, buyers should also look
for capabilities and features that support the entire software engineering lifecycle as well
as a wide user base that includes developers, data scientists, DBAs, IT operations and
business users.

A graph platform should include features for all of these users, offering extensions,
apps, libraries, APIs and SDKs, as well as integration with a variety of complementary
technologies.

18 neo4j.com
The Graph Technology Buyer’s Guide

Traditional RDBMS Developer Tooling

systems support the
ability to host and Technical user tools are essential for graph developers.

manage multiple In this area, the vendor should offer a visual programming environment that allows
databases within the developer to not only write queries, but to also see the graph upon which they are
working.
an instance of the
Developers need a variety of drivers and APIs (beyond just JDBC) with which to integrate
software. For graphs,
application-level functionality. Furthermore, the database should be embeddable within a
this is not yet the larger application as is often necessary in modern deployments.
case.
What to Look for: Developer Launchpad Application
An ideal developer experience is one that assembles all developer tools in one
application wrapper such that the developer always enjoys the same launchpad from
which to work with their graph database system.

Neo4j is the only graph database system to offer the Neo4j Desktop: an integrated
developer launchpad for graph projects. Neo4j Desktop provides access to the Enterprise
Edition of the Neo4j Graph Database along with its accompanying algorithm and
procedure libraries, as well as graph application examples. The Neo4j Desktop is free for
development use.

What to Look for: Visual Development Environment

A visual development environment should be included with the database to help
developers write clean, clear queries and also understand what those queries return.

Features of this environment should include keyword color-coding and auto-completed

values drawn from the database. It should also include a storyboard-like guide that allows
developers to teach and share step-by-step instructions. A visual graph drawing tool
should be included to help visualize the result set.

What to Look for: Application Drivers

Developers choose to write applications in a variety of programming languages, from Java,
Python or JavaScript to newer languages such as Go. The database should offer multiple
language-specific drivers to ease the burden for developers in designing systems that
include graph databases.

Neo4j offers commercial support for Java, JavaScript, .NET, Python and Go, while the
Neo4j community offers drivers for even more languages. TinkerPop also provides a
number of community-supported language integration drivers.

What to Look for: API Support

Programming interfaces are also important for feeding data into the database using
common protocols.

Interfaces such as Java or the database’s native language are desirable, as are interfaces
and exchange formats for other common occurrences. Any graph technology vendor
should also offer support for GraphQL (Facebook’s data exchange format) as a way of
interacting with applications above the database querying, as well as support for SPARQL,
the query language for RDF-based systems (Resource Description Framework), REST and
others.

19 neo4j.com
The Graph Technology Buyer’s Guide

Graph data isn’t rr Does the vendor support GraphQL?

typically “big data” in rr Does the vendor support SPARQL?
the classical sense, rr Does the vendor support REST APIs and frameworks?
and graph databases
rr Does the vendor actively support multiple emerging standards such as the
are not usually GRANDstack or similar?
swamped with the
size of indexes and What to Look for: Embeddability
JOIN tables in the Database embeddability is desirable for a number of reasons, including the ability for
same way as an OEMs to package the graph database along with application logic, user interfaces, cloud
containers like Kubernetes and more. More embeddability questions to consider include:
RDBMS.
rr Can the database be embedded within a surrounding application?
rr How easily can the graph system embed itself in larger applications?
rr Does the vendor support a robust OEM network and/or startup program?

Visualization Capabilities
Graph data visualization tools are extremely helpful in comprehending graph data
behaviors. These environments allow users to quickly see shapes within connected data
and determine where to look further.

What to Look for: Codeless Search, Business User Support

Business users benefit from a Google-like search and query experience, and a graph
visualization environment should therefore support natural language search functions.
In addition, visualization tools should help the user learn the graph as they use the
environment by offering built-in query suggestions and auto-completion.

Business user questions to consider include:

rr Is the graph data visualization tool codeless for non-technical users?

rr Does the visualization tool support natural language search?
rr Does the graph visualization tool offer built-in query suggestions?
rr Is the visualization tool web-based for ease of access?

What to Look for: Graph Exploration

A well-built graph visualization tool allows you to explore the graph in a variety of ways,
such as filtering nodes and relationships, lassoing parts of the graph, reviewing and
editing properties, highlighting paths, coloring nodes categories and including graphic
icons.

What to Look for: Linking & Embedding

A vendor’s graph visualization tool should allow users to link to views of the graph via URL
or external application for data illustrations on other mediums. These views should also
be fully embeddable for easy sharing and integration.

20 neo4j.com
The Graph Technology Buyer’s Guide

Database What to Look for: Developer Code Pasting & Viewing

administration is A robust graph visualization tool should supply fully-formatted query information to
greatly improved developers or other technical users so that they can adjust the first view or perspective
presented to other non-technical users.
with the inclusion
of administrative What to Look for: Graph Schema Perspectives for Data
tools not only for Security
DBAs, but also for
Any data visualization tool should include advanced security features that exploit the
programmers and graph’s ability to show different views of the data from the same graph.
analysts. Specifically, the tool should support presenting different views of the data or the
structure of the data (the schema of the graph) to different groups of users. Filters should
also have the option to be based on group-level security permissions.

What to Look for: Printing, Sharing & Storyboarding

Every graph visualization tool should offer multi-user sharing of graph views so that
a storyboard might be created to show multiple steps through a graph dataset.
Furthermore, the visualization tool should be able to print or otherwise present graph
content.

Deployment, Scaling & Administration

What to Look for: High Availability

Does the graph database platform offer a highly available solution designed to gracefully
survive one or more system failures? Often, this requires built-in application and data
redundancy. Some systems offer data redundancy upon a single, writable graph, and
then provide scaling for reads by cascading updates to replica systems deployed around
the world.

Other systems offer redundancy of multiple, duplicate writable graphs, which adds fault
tolerances to applications, networks and operations that lie above the storage system.
For these systems, it is important that their strength of consistency be carefully evaluated.
For graphs, consistency is vitally important, and when writing to multiple duplicates,
consistency must be coordinated carefully.

What to Look for: Write Scaling

Scaling graph writes is arguably the most difficult challenge faced by all graph database
vendors.

Current market expectations are that graph systems should horizontally scale for writes
as easily as they might for read optimization. The reality, however, is that creating a
horizontally scalable graph system requires the user to understand and plan their
graph schema carefully. For example, a user would need to design their schema so
that the graph could be partitioned in some form and therefore fit within any hardware
configuration.

Once this takes place, the user must then recognize how each partition will be identified
and how queries will be directed to the proper partition. Finally, the system will ultimately
need a means to leap queries from one partition to an adjacent one, which no graph
database product currently supports.

21 neo4j.com
The Graph Technology Buyer’s Guide

Even within the Due to every vendor’s inability to easily break apart a graph, write scaling must scale
vertically, requiring beefier hardware or a cloud environment. Such large, vertically scaled
almost wholly systems can scale writes in the following ways:
online world of
rr By offering more CPU-core backed sessions to accommodate concurrent
data technology, connections
face-to-face events,
rr By increasing RAM in which to cache most of the graph
gatherings and
conferences are rr By providing enough CPU power to minimize transaction cycle times
just as important What to Look for: Multi-Graph Support
since they establish
Some graph vendors have begun to solve the above-mentioned writing challenge. These
and strengthen vendors have identified means by which to isolate, view and operate upon one or many
relationships among graph clusters.
users, customers, Vendors are also identifying ways in which to extend application drivers to support
partners and directing queries to different graphs using a multi-graph routing table. While these graphs
are distinct, this is an illustration of the first stage in supporting a partitioned graph
employees.
solution.

What to Look for: Multi-Database & Multi-Tenancy

Traditional RDBMS systems support the ability to host and manage multiple databases
within an instance of the software. For graphs, this is not yet the case. Nearly all vendors
are single graph per software instance. Be sure to ask graph vendors these questions:

rr What does their product roadmap include for multiple databases?

rr Will their planned system support multiple tenants, hosting multiple databases?

What to Look for: Read Scaling

Read scaling is significantly easier for graphs than scaling writes.

Many graph database vendors support the ability to offer readable replicas of the main
graph. Some systems scale these replicas hierarchically. It is important to understand how
replicas stay consistent with the main writable-core instance to ensure that an application
or client can read their own writes (RYOW).

22 neo4j.com
The Graph Technology Buyer’s Guide

Now more than What to Look for: Caching Strategies

ever, enterprise Caching all or most of a graph in RAM offers a notable performance boost. In-memory
organizations are graphs significantly outperform those that are not cached.
deploying database Other questions a vendor should be able to answer about their graph-caching strategies
instances to the include:
cloud in addition to rr If only sections of a graph fit in memory, then what functions does the vendor offer?
on-premises. rr Can a language driver point queries to pre-cached information?
rr Can the system reinstate or reheat the cache after system restart?

What to Look for: Scaling Data Volumes

Graph data isn’t typically “big data” in the classical sense, and graph databases are not
usually swamped with the size of indexes and JOIN tables in the same way as an RDBMS.
In fact, both JOIN tables and large indexes fall away when moving your data into a native
graph database.

Scaling large volumes of data is accomplished via big data processing technologies such
as Apache Spark for analytic activities or Apache Kafka for data streaming.

Ask vendors about their ability to execute graph queries upon Apache Spark, or whether
they can materialize graphs from Spark DataFrames. Look for mature clustering
technologies that use tools such as the Raft protocol instead of a simpler master-slave
architecture.

What to Look for: Cloud Support

Now more than ever, enterprise organizations are deploying database instances to
the cloud in addition to on-premises. Ask these questions when considering a graph
database vendor who claims to have cloud support:

rr Does the vendor offer both on-premises and cloud-based versions of its software?
rr Does the vendor offer support for multiple cloud vendors beyond Amazon Web
Services, such as Microsoft Azure, Google Cloud Platform or Alibaba Cloud?

rr Are pricing and consumption costs transparent to the customer?

rr Does the cloud offering mirror the vendors’ on-premises functionality?

What to Look for: DBaaS Support

Analysts have predicted that a fully functional Graph Database-as-a-Service will become
very popular in the years to come. Ask vendors these questions in relation to their DBaaS
plans:

rr What is the vendor’s plan to offer this type of functionality?

rr What subscription tiers will they offer?
rr Is there a multi-tenant offering allowing a value-added software vendor to host
multiple downstream customers?

23 neo4j.com
The Graph Technology Buyer’s Guide

What to Look for: Container Support & Management

Using containers benefits the organization in multiple ways, including simplified
configurations, easier upgrade transitions and scaled replicas.

rr What container-based operating environment is supported by the graph system –

Docker, Kubernetes or others?

rr Is this a common deployment model for the vendor?

What to Look for: Security

Database integrity and security are extremely important in enterprise data management
systems. Ask vendors about these most common database security features and
integrations:

rr How does the environment secure users’ authentication and authorization rights?
rr Does the system integrate with popular directories such as LDAP or Active
Directory?

rr Are security groups supported?

rr Is Kerberos supported?
rr Are client-server traffic and server-to-server traffic encrypted inside of the cluster?
rr Is event security logging supported and can administrators disable and re-enable
users?

rr Is security transparent to the developer, or must the developer themselves manage

security within the applications they build?

rr Is the data encrypted on disk? (This is often done at the operating system and file
system level.)

rr Is the data encrypted in transit in server-to-server activities, in administrative

functions and to users’ browsers?

What to Look for: Database Administration

Database administration is greatly improved with the inclusion of administrative tools not
only for DBAs, but also for programmers and analysts.

rr Does the system offer the ability to check logging activity or query executions?
rr Can users terminate runaway queries?
rr Can users view query execution plans?
rr Can administrators review usage statistics that may be used for billing or
departmental chargeback?

24 neo4j.com
The Graph Technology Buyer’s Guide

Business Model, Focus & Staying Power

Many vendors offer graph extensions and add-ons, which may not provide the optimal
graph experience. They may also unintentionally introduce corruptions, as detailed in the
earlier section on native graph storage.

rr To what degree is the vendor focused specifically on graph support?

rr What level of operations and activity are dedicated to graph systems?
Native graph systems and vendors have and will dedicate significantly more resources to
the success of graphs than multi-model vendors, whose attention will be distracted by
the other database models they claim to support.

What to Look for: Partnerships

Ask who the vendor considers to be their key partners in the market. Partnerships should
include cloud vendors, consulting and implementation providers, software OEMs, value-
added resellers, territorial resellers and education centers.

rr Does the vendor have a wide variety of partnerships throughout both business and
technology landscapes?

rr How closely does the vendor manage and work with their partners? Are they
partners in name only?

What to Look for: Ecosystem

You should also consider a vendor’s customer and user ecosystem. Questions that you
should ask a vendor include:

rr How large, vocal and active is the vendor’s ecosystem?

rr What are their most frequent questions, issues and ideas?
rr Where do they share ideas and information – a community forum, news group or
developer website?

What to Look for: Events & Conferences

Even within the almost wholly online world of data technology, face-to-face events,
gatherings and conferences are just as important since they establish and strengthen
relationships among users, customers, partners and employees. Questions you should
ask a vendor include:

rr What events does the vendor host to develop its community?

rr How many regular meetups does it participate in?
rr What breadth of geography does a vendor cover in their events and in-person
outreach?

rr Does the vendor hold a regular conference dedicated to graph technology?

25 neo4j.com
The Graph Technology Buyer’s Guide

What to Look for: Customer Feedback

Any vendor can claim to have top-name customers who are happy with their product,
but not every vendor can back up those claims with solid references and case studies.
Questions you should ask a graph technology vendor include:

rr Who are the vendor’s referenceable customers?

rr Do they have a public listing of customers available on their website?
rr Will the vendor’s sales team allow you to speak to some of their current customers?
Beware of signs that may indicate a conflict of interest, such as a corporate investor also
posing as a customer reference.

What to Look for: Licensing Model

Not all customers’ software needs are the same. A mature vendor offers a variety of
licensing models for their graph database software to match the needs of different
customer requirements. Questions that you should ask a vendor include:

rr Does the vendor offer a variety of license models to suit the buyer’s preferred
deployment and feature combinations?

rr Does the vendor publish their price list?

What to Look for: Mature Technology vs. Technology Past its

Prime
Database analyst Curt Monash famously observed that “the first rule of databases is that
it takes at least five to seven years to build a good, stable one. The second rule of
databases is that there is no exception to rule number 1.” This maturity requirement rule
is intended to reinforce the importance of most of the earlier requirements.

Likewise, look out for aging technology that is losing its place in the market. This is often
referred to as the “Goldilocks effect,” whereby mature technologies grow stale, paving
the way for newer ones. The aging technology is then slowly put to pasture as a cash
cow or retrofitted with bolt-on updates that simply add features with little regard to their
effectiveness. The result is a Swiss Army Knife – a tool that does a little of everything but
none of it well.

This phenomenon is currently happening in the multi-model database space, especially

for graphs. Graph add-ons are becoming the saw in a Swiss Army Knife – you can’t build a
robust house with it.

What to Look for: Staying Power & Customer Support to Mini-

mize Technology Risk
The staying power requirement acts as a reminder that software must be maintained and
enhanced, or else it will atrophy and cease to keep pace. Beware of small software teams
or teams of consultants charged with writing and maintaining the software. Additionally,
be cautious of companies nested inside of companies serving as embedded subsidiaries
of a conglomerate.

In addition, a mature graph technology vendor should have the resources necessary
to support your graph database deployment should any issue or problems arise. No
technology is ever deployed perfectly (no matter whether the vendor or the buyer are
at fault), but given the newness of the graph paradigm, a vendor must have a dedicated
team of support engineers who can help with your critical issues or questions as they
arise. This team should be geographically spread out and available for emergency
support around the clock.

26 neo4j.com
The Graph Technology Buyer’s Guide

As the graph Below are questions to ask vendors:

technology market rr Is the vendor well-backed financially?

continues to rr Does the vendor operate as a fiscally responsible business?
grow and mature,
rr Is the vendor’s product team properly staffed and supported?
purchasing decisions
rr Does the vendor have a team of professional support engineers to help with critical
may become more issues around the globe and around the clock?
clear as vendors
coalesce around
shared standards Conclusion
(such as query
Making an informed buying decision for graph technology isn’t always easy.
languages, data
Because the graph paradigm is still such a new way of working with data, purchasing
formats, etc.). teams are often held back by RDBMS norms when considering how to vet a graph
database vendor. But as has been shown above, factors such as data consistency,
performance, scalability and other important characteristics all have different applications
when it comes to graph databases versus their relational counterparts.

As the graph technology market continues to grow and mature, purchasing decisions
may become more clear as vendors coalesce around shared standards (such as query
languages, data formats, etc.). But until then, we hope that this buyer’s guide has been
helpful in clarifying which factors are most important when considering a graph database
product – and which factors might just be marketing fluff with no substance behind them.

For your convenience, we’ve included a blank checklist in Appendix A that lists all of the
factors outlined in the sections above.

If you’d like to talk to a graph expert about the advantages of the Neo4j Graph Platform,
please contact us today.

27 neo4j.com
The Graph Technology Buyer’s Guide

This Appendix includes a blank checklist for the reader in order to independently evaluate and compare graph database choices.

Appendix A
The Buying Criteria for Graph Technology
Open Source Foundation & Community
rr Is the graph database built on an open source foundation?
rr Does the graph database vendor have an active open source community?
Native Graph Storage
rr Transposition vs. Persisted Relationships
rr Is the offering a native or non-native graph database?
rr Is there a graph-to-document or graph-to-column transposition step buried in the software?
rr Index-Free Adjacency
rr Does the graph database support index-free adjacency for high-performance graph traversals across deep graph
datasets?

ACID Compliance
rr Durability
rr Is the graph database systems rigorously tested for data durability?
rr Does the graph database lose data during systemic hardware failures?
rr Consistency Checks
rr Does the database offer any level of consistency checking for graphs?
rr What safety measures are available to ensure writes are written correctly?
rr Do these measures work as expected in a clustered, High Availability (HA) environment?
rr Does the database vendor’s documentation contain warnings, such as beware of ghost vertices or floating or
untethered edges?

Graph Query Languages

rr Declarative vs. Imperative Query Language
rr Does the graph database system use a declarative query language, imperative query language or a mixture of both?
rr Compiled vs. Interpreted Instructions
rr Are queries compiled into binary packages prior to execution within the database (optimization for a single query)?
rr Or, are queries interpreted by the database in a way that allows for ad-hoc querying and changes on the fly
(optimization for user agility)?

rr Graph Query Language Standardization

rr Is the database vendor part of the ISO W3C standardization movement around Graph Query Language (GQL)?
rr Does the vendor have plans to be compatible with GQL in the future?

28 neo4j.com
The Graph Technology Buyer’s Guide

Hybrid Transactional-Analytic Platforms (HTAP)

rr OLTP Applications
rr Does the graph database vendor have verified example implementations and customer references for a variety of OLTP
use cases?

rr Does the graph database support geospatial data types?

rr Does the graph system support temporal formats and calculations?
rr Does the graph system support mapping, indexing and searching for specific text within the graph? (full-text search and
indexing)

rr Query Traversal Performance Benchmark Comparisons

rr If the vendor has published a benchmark comparison, does it provide verifiable and reproducible tests that include
data, documentation, configuration and execution methods as well as results?

rr Did the vendor use the Linked Data Benchmark Council (LDBC) array of tests?
rr Did the vendor work with independent third parties that have performed other vendor-neutral comparisons?
rr Does database performance deteriorate exponentially when the number of traversal hops across the graph increases?
rr Does the benchmark compare a pre-compiled binary query against an interpreted query?
rr Does the vendor fully exercise the system or simply highlight its most favorable tasks and results?
rr For example, are workloads mixed as both transactional writes and analytic reads, or is only one type of activity
executed?

rr Are caching strategies equal?

rr For example, are both graphs fully loaded into memory, or is only one product fully loaded into memory while the
other swaps between cache and disk?

rr Are one-time data loading operations overly emphasized – versus multi-user or multi-session tests – which better
reflect regular usage?

rr Are benchmark sessions multi-user, or are they single sessions that utilize all available CPU resources to parallelize
query execution?

rr Does the benchmark consider real-world, mixed read-write workloads?

rr Are the systems configured optimally and fairly along criteria such as system sizing, memory and/or tuning config
parameters?

rr Does the benchmark only use database defaults, rather than being optimized for performance at scale?
rr Graph Analytics (OLAP Applications)
rr Has the vendor published information about how to perform graph analytics exercises using graph algorithms?
rr Does the vendor offer support for popular graph algorithms?
rr Does the vendor include documentation explaining the appropriate use and application of various graph algorithms?
rr Data Integration & Ingestion
rr Does the graph system operate within the Apache Spark ecosystem?
rr Can graphs be materialized from Spark DataFrames?
rr Can graph structures be directly imported into the graph database system?
rr Can graph algorithms be run against data in Apache Spark?
rr Does the vendor offer libraries, SDKs and data integration adapters to popular data management tools?

29 neo4j.com
The Graph Technology Buyer’s Guide

rr Open source tools: Talend and Kettle

rr Commercial tools: Informatica, Ab Initio, iWay, Trifacta and Tamr
rr Does the graph database support CSV data import?
rr Does the graph database offer high-speed and bulk data ingestion?
rr Does the graph solution include the ability to defer index building and consistency checking until after the data is
ingested?

Graph Platform with Tools & Support for All Types of Users
rr Developer Tooling
rr Does the platform include a launchpad application wrapper for all developer tools related to the graph database
system?

rr Does the platform offer a visual development environment to help developers write clean, clear queries and
understand what those queries return?

rr Does the visual development tool include keyword color-coding and auto-completed values drawn from the
database?

rr Does the platform offer commercial support for multiple language drivers for the most popular programming
languages?

rr Does the vendor’s wider user community offer drivers for other popular programming languages?
rr Does the graph platform offer API support for Java or the database’s native language?
rr Does the platform include support for GraphQL?
rr Does the platform support integration with SPARQL?
rr Does the platform support REST APIs and frameworks?
rr Does the vendor actively support multiple emerging standards such as the GRANDstack or similar?
rr Can the database be embedded within a surrounding application?
rr How easily can the graph system embed itself in larger applications?
rr Does the vendor support a robust OEM network and/or startup program?
rr Data Visualization Capabilities
rr Does the platform include a native graph data visualization tool?
rr Is the graph data visualization tool codeless for non-technical users?
rr Does the visualization tool support natural language search?
rr Does the graph visualization tool offer built-in query suggestions?
rr Is the visualization tool web-based for ease of access?
rr Does the visualization tool allow non-technical users to easily navigate and explore a given graph dataset?

rr Does the graph visualization tool allow users to link to (or embed) views of the graph?
rr Does the visualization tool let technical users execute Cypher queries or adjust perspectives for other non-technical
users?

rr Does the data visualization tool include advanced security features such as user- or group-level viewing permissions?
rr Does the visualization tool offer multi-user sharing of graph views, storyboarding, printing or other presentation
methods?

30 neo4j.com
The Graph Technology Buyer’s Guide

rr Deployment, Scaling & Administration

rr Does the graph database platform offer a highly available solution designed to gracefully survive one or more system
failures?

rr Does the graph platform vertically scale for your expected write load?
rr Does the graph platform include multi-graph support for both writes and reads?
rr Are multi-database features included in the vendor’s product roadmap?
rr Will the vendor’s planned system support multiple tenants, hosting multiple databases?
rr Does the graph platform support robust read scaling?
rr Does the graph database offer readable replicas of the main graph?
rr How do replicas stay consistent with the main writable-core instance to ensure that an application or client can read
their own writes (RYOW)?

rr What are the vendor’s graph-caching strategies?

rr If only sections of a graph fit in memory, then what functions does the vendor offer?
rr Can a language driver point queries to pre-cached information?
rr Can the system reinstate or reheat the cache after system restart?
rr Can the graph platform execute graph queries upon Apache Spark?
rr Can the graph platform materialize graphs from Spark DataFrames?
rr Does the graph platform utilize mature clustering technologies such as the Raft protocol, or does it rely on a simpler master-
slave architecture?

rr Does the vendor offer both on-premises and cloud-based versions of its software?
rr Does the vendor offer support for multiple cloud vendors beyond Amazon Web Services, such as Microsoft Azure,
Google Cloud Platform or Alibaba Cloud?

rr Are pricing and consumption costs transparent to the customer?

rr Does the cloud offering mirror the vendors’ on-premises functionality?
rr Does the vendor plan to offer a Graph Database-as-a-Service (DBaaS) in the future?
rr What is the vendor’s plan to offer this type of functionality?
rr What subscription tiers will they offer?
rr Is there a multi-tenant offering allowing a value-added software vendor to host multiple downstream customers?
rr What container-based operating environment is supported by the graph system: Docker, Kubernetes or others?
rr Is this a common deployment model for the vendor?
rr What are the graph vendor’s most common database security features and integrations?
rr How does the environment secure users’ authentication and authorization rights?
rr Does the system integrate with popular directories such as LDAP or Active Directory?
rr Are security groups supported?
rr Is Kerberos supported?
rr Are client-server traffic and server-to-server traffic encrypted inside of the cluster?
rr Is event security logging supported?
rr Can administrators disable and re-enable users?

31 neo4j.com
The Graph Technology Buyer’s Guide

rr Is security transparent to the developer, or must the developer themselves manage security within the applications they
build?

rr Is the data encrypted on disk?

rr Is the data encrypted in transit in server-to-server activities, in administrative functions and to users’ browsers?
rr What database administrative tools are included in the graph platform for DBAs, programmers and analysts?
rr Does the system offer the ability to check logging activity or query executions?
rr Can users terminate runaway queries?
rr Can users view query execution plans?
rr Can administrators review usage statistics that may be used for billing or departmental chargeback?
Business Model, Focus & Staying Power
rr To what degree is the vendor focused specifically on graph support?
rr What level of operations and activity are dedicated to graph systems?
rr Who are the vendor’s key partners in the market?
rr Does the vendor have a wide variety of partnerships throughout both business and technology landscapes?
rr Cloud vendors?
rr Consulting and implementation providers?
rr Software OEMs?
rr Value-added resellers?
rr Territorial resellers?
rr Training and education centers?
rr How closely does the vendor manage and work with their partners?
rr Are they partners in name only?
rr What is the vendor’s customer and user ecosystem like?
rr How large, vocal and active is the vendor’s ecosystem?
rr What are their most frequent questions, issues and ideas?
rr Where do they share ideas and information – a community forum, news group or developer website?

rr What sorts of face-to-face events and conferences does the vendor participate in?
rr What events does the vendor host to develop its community?
rr How many regular meetups does it participate in?
rr What breadth of geography does a vendor cover in their events and in-person outreach?
rr Does the vendor hold a regular conference dedicated to graph technology?
rr Who are the vendor’s reference customers?
rr Do they have a listing of public customers available on their website?
rr Will the vendor’s sales team allow you to speak to some of their current customers?
rr Do any of the vendor’s reference customers hold a conflict of interest, such as also being a corporate investor?

32 neo4j.com
The Graph Technology Buyer’s Guide

rr What is the vendor’s licensing model for their graph database software?
rr Does the vendor offer a variety of license models to suit the buyer’s preferred deployment and feature combinations?
rr Does the vendor publish their price list?
rr How long has the vendor been working on their graph database product?
rr Does is meet the Monash Rule of 5-7 years of minimum development for a stable database product?
rr Are the graph capabilities merely an add-on of an older, legacy product?
rr Is the graph technology vendor have the business staying power to build, update and support the product for years to
come?

rr Is the vendor well-backed financially?

rr Does the vendor operate as a fiscally responsible business?
rr Is the vendor’s product team properly staffed and supported?

rr Does the vendor have a team of professional support engineers to help with critical issues around the globe and
around the clock?

Neo4j is the leading graph database platform that drives innovation and competitive advantage at Airbus, Comcast, Questions about Neo4j?
eBay, NASA, UBS, Walmart and more. Hundreds of thousands of community deployments and more than 300
customers harness connected data with Neo4j to reveal how people, processes, locations and systems are
Contact us around the
interrelated.
globe:
Using this relationships-first approach, applications built using Neo4j tackle connected data challenges including info@neo4j.com
artificial intelligence, fraud detection, real-time recommendations and master data. Find out more at Neo4j.com. neo4j.com/contact-us

© 2019 Neo4j. All rights reserved. Front cover image: Anna Sullivan on Unsplash. neo4j.com