Change Password

Input error
Input error
Input error
Submit

ULTIPA GRAPH Benchmark Report (2020)

This document serves as a product performance and capability benchmark report. Its main contents cover performance comparison matrix between Ultipa Graph, a graph compute engine and graph database product, and other peer products (including but not limited to Neo4J, Janus Graph, Tigergraph, and Amazon Neptune).

Part 0: Testing Bed & Preparation

To ensure that the performance benchmark is both fair and easy to digest, we first need to set up the testing bed, the hardware platform, which we settled on Intel Architecture PC servers.

Due to the license restriction of certain graph databases, together with the data volume, we settled on a fairly standard hardware setup, please refer to Table 1 for configuration details.

In Part 5, Scalability was tested against Ultipa Graph & Janus Graph only due to the unavailability of cluster-edition of the other graph databases.

Metrics

Configurations

Remarks

HW Vendor

Unknown

Aliyun ECS

CPU

Intel X86-64 2.0G

32 vCPUs

Memory

244GB

Configurations per needs.

Max Bandwidth

5 Gbps

2nd SSD attached over network.

Disk

80+80 SSD (over cloud)

Attached and Cloud-Disk

OS

Ubuntu Linux

18.04.x LTS 64-bit

Table 1: Cloud-Based Hardware Configurations

Note, the above ECS configuration is computationally 20-30% less capable to the AWS cloud hardware setup utilized by vendors like TigerGraph and others. But, we are okay to carry on with this apple-to-apple comparison.

AWS Metrics

Configurations

Remarks

HW Vendor

AWS EC2 Machine

r4.8xlarge

vCPUs

Intel X86-64bit 2.0G+

32vCPUs

Memory

244 GB

Standard EC2 RAM offering.

Disk

160 GB

SSD (Cloud Disk)

Max Bandwidth

7 Gbps

Irrelevant for non-scalability.

IOPS

60,000

Cloud SSD

OS

Ubuntu Linux

18.04.x LTS 64-bit

Table 2: Public Cloud (AWS) Hardware Configurations by other vendors.

Note: Other vendors refer to TigerGraph, Neo4J, JanusGraph, RedisGraph, and others.

Metrics

Configurations

Remarks

Java Edition

1.8

 

Worker Threads

32

# of threads (vCPU)

Heap Size

16GB (out of 20GB system)

For smaller dataset (see below)

Memory Allocation

200GB

For larger dataset (see below)

Table 2-2: JVM Configurations for Java-based Systems (i.e., Neo4J, JanusGraph)

Want to read more?