Performance test report-PolarDB(PolarDB)-阿里云帮助中心

This topic describes the performance test results of the document database (compatible with the MongoDB protocol) of PolarDB for MySQL based on Yahoo! Cloud Serving Benchmark (YCSB), including QPS and latency metrics across scenarios such as data loading, read-only, mixed read/write, and write-only.

Test environment

The following table lists the configurations used in this performance test.

Item	Configuration
PolarDB cluster	The engine is MySQL 8.0.2, the cluster edition series, and the cluster specification is `polar.mysql.x8.xlarge` (8 vCPUs, 64 GB memory). The storage type is PSL5.
Test client	An Alibaba Cloud ECS instance with the specification `ecs.c6e.2xlarge` (8 vCPUs, 16 GB memory), deployed in the same VPC as the PolarDB cluster.

Test tool

YCSB (Yahoo! Cloud Serving Benchmark) is used as the benchmark tool for this test. YCSB is a widely used key-value store performance testing framework that supports multiple database engines and can simulate read and write workloads in real-world scenarios.

Each document in the test data model contains 10 fields, with each field being 100 bytes in size. The size of a single document is approximately 1 KB.

Performance tests

Test scenarios

This performance test covers the following four scenarios.

Scenario	Read/Write ratio	Description
Load	Write-only	Bulk load test data into the database.
Read-only	Read:Write = 100:0	All requests are read operations. This scenario evaluates the throughput and latency in a read-only workload.
Read/Write mix	Read:Write = 50:50	Read and write operations each account for 50%. This scenario evaluates the performance under a mixed workload.
Write-only	Read:Write = 0:100	All requests are write operations. This scenario evaluates the throughput and latency in a write-only workload.

Test metrics

The following metrics are used to measure performance in this test.

Metric	Description
QPS	The number of requests processed per second (Queries Per Second). A higher value indicates greater throughput.
RAL (ms)	Read average latency, in milliseconds. A lower value indicates faster read operation response.
WAL (ms)	Write average latency, in milliseconds. A lower value indicates faster write operation response.

Test results

The following table shows the performance test results under different thread counts and scenarios.

Scenario	Threads	QPS	RAL (ms)	WAL (ms)
Load	32	21,020.8	-	1.94
Read-only (r:w=100:0)	16	56,722.7	0.27	-
Read-only (r:w=100:0)	32	115,369.5	0.27	-
Read/Write mix (r:w=50:50)	16	29,225.7	0.32	0.76
Read/Write mix (r:w=50:50)	32	41,533.0	0.39	1.14
Write-only (r:w=0:100)	16	18,455.8	-	0.86
Write-only (r:w=0:100)	32	25,791.0	-	1.23

Note

In the read-only scenario, QPS increases nearly linearly with the number of threads, reaching 115,369.5 at 32 threads. Read latency remains stable at 0.27 ms.
In the mixed read/write scenario, both read and write latencies remain at low levels, demonstrating balanced overall performance.
In the write-only scenario, QPS reaches 25,791.0 at 32 threads, with a write latency of 1.23 ms.

Test command reference

The following are example commands used in the YCSB tests. You can adjust the parameter values as needed.

Load data

./bin/ycsb load mongodb -s \
  -p workload=site.ycsb.workloads.CoreWorkload \
  -p recordcount=10000000 \
  -p mongodb.url="mongodb://<user>:<password>@<endpoint>:<port>/<db>" \
  -p core_workload_insertion_retry_limit=3 \
  -threads <thread_count>

Run tests

  ./bin/ycsb run mongodb -s \
  -p workload=site.ycsb.workloads.CoreWorkload \
  -p recordcount=10000000 \
  -p operationcount=10000000 \
  -p insertproportion=0 \
  -p readproportion=<read_ratio> \
  -p updateproportion=<write_ratio> \
  -p mongodb.url="mongodb://<user>:<password>@<endpoint>:<port>/<db>" \
  -threads <thread_count>

The following table describes the parameters in the test commands.

Parameter	Description
`mongodb.url`	The connection address of the document database. Replace `<username>`, `<password>`, `<endpoint>`, and `<port>` with the actual account and access address of the document database.
`recordcount`	The total number of records in the load phase.
`operationcount`	The total number of operations in the run phase.
`readproportion`	The proportion of read operations. Set to 100 for the read-only scenario, 0 for the write-only scenario, and 50 for the mixed read/write scenario.
`updateproportion`	The proportion of write operations. Set to 100 for the write-only scenario, 0 for the read-only scenario, and 50 for the mixed read/write scenario.
`threads`	The number of concurrent threads. In this test, 16 and 32 threads are used.