Sysbench benchmark report of MonoSQL
Test environment (AWS EC2/Auto Scaling Group)
Hardware configuration
| Service type | EC2 type | Instance count |
|---|---|---|
| MonoSQL | c5.2xlarge | 2-16 |
| Sysbench | c5.9xlarge | 1 |
Software version
| Service type | Software version |
|---|---|
| MonoSQL | 1.3.1 |
| Sysbench | 1.0.20_2 |
| OS | Ubuntu20.04 |
Parameter configuration
MonoSQL parameter configuration
plugin_maturity=experimental
datadir=/home/ubuntu/data0
max_connections=1200
plugin_dir=/home/ubuntu/install/lib/plugin
max_connect_errors=2147483647
performance_schema=ON
skip-log-bin
port=3306
socket=/tmp/mysqld3306.sock
plugin_load_add=ha_monosql
monosql
monosql_dynamodb_default_credentials=on
monosql_dynamodb_region=us-east-1
monosql_keyspace_name=mono
monosql_sqs_queue_name=monosql-queue
Prepare test data
Run the following command to prepare the test data. Note that table number is set 100 due to the DynamoDB on-demand table's initial throughput limit is 3000. If you use provisioned table with high RCU and WCU, table number could be set to 1 and table size can be set to 10 milltion or more.
sysbench /usr/share/sysbench/oltp_common.lua --mysql_storage_engine=monosql --tables=100 --table_size=10000 --mysql-user=sysb --mysql-password=sysb --mysql-host=$host --mysql-port=3306 --mysql-db=sbtest --time=600 --threads=240 --report-interval=10 --auto_inc=off --create_secondary=false --mysql-ignore-errors=all prepare
Run the test
Run the following command to perform the test:
$testname's value includes: oltp_point_select/oltp_update_non_index/oltp_insert
sysbench $testname --mysql_storage_engine=monosql --tables=100 --table_size=10000 --mysql-user=sysb --mysql-password=sysb --mysql-host=$host --mysql-port=3306 --mysql-db=sbtest --time=600 --threads=$thr_num --report-interval=10 --auto_inc=off --create_secondary=false --mysql-ignore-errors=all run
The following performance result shows that MonoSQL has a linear scalibility powered by DynamoDB. By adding more MonoSQL instances and more database connections, you can get single-digit millisecond latency at scale.
OLTP_POINT_SELECT Performance
| Test Type | Thread Num | TPS | Avg Latency | 95th percentile |
|---|---|---|---|---|
| c5.2xlarge * 2 | 240 | 30338.69 | 7.9 | 12.52 |
| c5.2xlarge * 4 | 480 | 61920.87 | 7.7 | 12.30 |
| c5.2xlarge * 8 | 960 | 126394.36 | 7.5 | 11.87 |
| c5.2xlarge * 16 | 1920 | 249806.74 | 7.6 | 12.30 |
OLTP_UPADTE_NON_INDEX Performance
| Test Type | Thread Num | TPS | Avg Latency | 95th percentile |
|---|---|---|---|---|
| c5.2xlarge * 2 | 240 | 29365.00 | 8.1 | 11.87 |
| c5.2xlarge * 4 | 480 | 59023.22 | 8.1 | 11.87 |
| c5.2xlarge * 8 | 960 | 119305.26 | 8.0 | 11.45 |
| c5.2xlarge * 16 | 1920 | 232533.59 | 8.2 | 11.87 |
OLTP_INSERT Performance
| Test Type | Thread Num | TPS | Avg Latency | 95th percentile |
|---|---|---|---|---|
| c5.2xlarge * 2 | 240 | 29564.20 | 8.1 | 11.87 |
| c5.2xlarge * 4 | 480 | 59280.73 | 8.1 | 11.87 |
| c5.2xlarge * 8 | 960 | 119394.46 | 8.0 | 11.45 |
| c5.2xlarge * 16 | 1920 | 233812.24 | 8.2 | 11.87 |
