Performance Test Methods

Test Environment

• Region: AP-Singapore
• AZ: AZ1, AZ2, and AZ3 (three-AZ deployment)
• Elastic Cloud Server (ECS): h3.4xlarge.2 with 16 vCPUs, 32 GB of memory, and CentOS 7.5 64-bit image
• Nodes per instance: 3
• Instance specifications: All specifications described in Table 1

  Table 1 Instance specifications

  No.	Specifications
  Cluster 1	4 vCPUs | 16 GB
  Cluster 2	8 vCPUs | 32 GB
  Cluster 3	16 vCPUs | 64 GB
  Cluster 4	32 vCPUs | 128 GB

Test Tool

YCSB is an open-source tool for testing performance of databases. In this test, YCSB 0.15.0 is used. RoundRobinPolicy is used for load balancing.

For details on how to use this tool, see YCSB.

Test Metrics

Operations per Second (OPS): operations executed by a database per second

Test Procedure

1. Configure the workload file.

   Set values for fields readproportion, insertproportion, updateproportion, scanproportion, and readmodifywriteproportion in the file by referring to Table 2.

   Set a value for field recordcount in the file by referring to Table 3.

2. Use workload-insert-mostly as an example. Run the following command to prepare test data:

   sh bin/ycsb.sh load cassandra-cql -P workloads/workload-insert-mostly -p "hosts=${ContactPoints}" -p "port=${port}" -p "cassandra.username=${username}" -p "cassandra.password=${password}" -p operationcount=400000000 -p recordcount=400000000 -p exportfile=./data_load.exp -threads ${threadNum} -s > data_load.log 2>&1 &

3. Use workload-insert-mostly as an example. Run the following command to test performance:

   sh bin/ycsb.sh run cassandra-cql -P workloads/workload-insert-mostly -p "hosts=${ContactPoints}" -p "port=${port}" -p "cassandra.username=${username}" -p "cassandra.password=${password}" -p operationcount=9000000 -p recordcount=9000000 -p maxexecutiontime=3600 -p exportfile=./workload-insert-mostly.exp -threads ${threadNum} -s > workload-insert-mostly_run.log 2>&1 &

Test Models

• Workload model

  Table 2 Workload models

  Workload Model	Description
  More read requests than write requests workload-read-mostly	95% read, 5% update
  Balanced read and write requests workload-read-write-combination	50% update, 50% read
  Balanced read and rewrite requests workload-read-modify-write	50% read, 50% read-modify-write
  Read, update, and write requests workload-mixed-operational-analytical	65% read, 25% update, 10% insert
  More write requests than read requests workload-insert-mostly	90% insert, 10% read

• Data model

  fieldlength = 100, fieldcount = 10

• Preset data volume

  Different preset data volumes were used to test performance of instances of each type of specifications.

  The following table describes the preset data volumes.

  Table 3 Preset data volumes

  No.	Specifications	Preset Data Volume
  Cluster 1	4 vCPUs | 16 GB	50 GB
  Cluster 2	8 vCPUs | 32 GB	100 GB
  Cluster 3	16 vCPUs | 64 GB	200 GB
  Cluster 4	32 vCPUs | 128 GB	400 GB