This topic describes the design, procedure, and results of a change data capture (CDC) node performance test for PolarDB-X. Use it as a reference to evaluate CDC node performance in your own instance.
Background
CDC nodes act as log nodes in a PolarDB-X instance, providing a binary logging service compatible with MySQL binary logs. Subscribe to and consume binary logs from CDC nodes to meet requirements across different business scenarios.
This topic uses TPC Benchmark C (TPC-C), Sysbench, and Large Transaction to evaluate CDC node performance. The following metrics measure performance. For details, see Log resource monitoring.
| Metric | Abbreviation | Description | When to pay attention |
|---|---|---|---|
| Delay Time | DT | Latency for processing binary log events. Unit: milliseconds. | Sustained increases indicate that CDC nodes are falling behind the upstream write rate. |
| Bytes Per Second | BPS | Data size of binary log events processed per second. Unit: byte/s. | Drops during large-transaction workloads indicate that disk I/O is involved. See Large Transaction results. |
| Events Per Second | EPS | Number of binary log events processed per second. | — |
| Transactions Per Second | TPS | Number of transactions processed per second. All transactions in binary logs are counted. | — |
Test design
Instance specifications
| Node type | Specifications | Number of nodes |
|---|---|---|
| Compute node | 32 vCPUs, 128 GB of memory | 12 |
| Data node | 32 vCPUs, 128 GB of memory | 16 |
| CDC node | 16 vCPUs, 32 GB of memory | 2 |
Stress test environment
ECS instance type: ecs.hfc6.16xlarge (64 vCPUs, 128 GB of memory)
To generate large volumes of traffic, you may need more than one Elastic Compute Service (ECS) instance.
Procedure
TPC-C
Follow the procedure in TPC-C test with the following settings:
-
In
props.mysql, setwarehousesto2000. -
In
props.mysql, setrunMinsto10. -
In
runBenchmark.sh, set the Java Virtual Machine (JVM) heap size to-Xms60g -Xmx60g.
Sysbench
Follow the procedure in Sysbench test. Adjust the number of concurrent threads to generate different traffic volumes.
oltp_write_only
sysbench --config-file='sysb.conf' --db-ps-mode='disable' --skip-trx='off' \
--mysql-ignore-errors='all' --tables='16' --table-size='10000000' \
--threads={Number of concurrent threads} oltp_write_only runoltp_insert
sysbench --config-file='sysb.conf' --db-ps-mode='disable' --skip-trx='on' \
--mysql-ignore-errors='all' --tables='16' --table-size='10000000' \
--threads={Number of concurrent threads} oltp_insert runLarge Transaction
Download large_transaction.sql to construct transactions of varying sizes. Inserting 200,000 rows produces a transaction of 500 MB.
Test results
TPC-C
| tpmC | DT (ms) | BPS (MB/s) | EPS | TPS |
|---|---|---|---|---|
| 260,293.98 | 60 | 65 | 307,403 | 9,078 |
| 515,425.45 | 80 | 128 | 601,757 | 17,980 |
| 783,525.69 | 100 | 197 | 956,529 | 27,935 |
| 1,087,352.38 | 120 | 270 | 1,292,464 | 38,867 |
| 1,283,716.12 | 150 | 325 | 1,549,433 | 45,968 |
| 1,500,365.16 | 220 | 380 | 1,810,926 | 53,592 |
Sysbench
oltp_write_only
| QPS | DT (ms) | BPS (MB/s) | EPS | TPS |
|---|---|---|---|---|
| 106,081.68 | 45 | 34 | 274,068 | 17,761 |
| 215,321.43 | 60 | 68 | 532,190 | 35,672 |
| 302,061.64 | 100 | 97 | 788,396 | 50,560 |
| 408,594.41 | 287 | 128 | 1,023,873 | 68,702 |
| 518,068.92 | 420 | 152 | 1,201,216 | 84,654 |
oltp_insert
| QPS | DT (ms) | BPS (MB/s) | EPS | TPS |
|---|---|---|---|---|
| 50,984.67 | 46 | 25 | 309,384 | 50,984.67 |
| 106,907.38 | 95 | 51 | 645,040 | 106,907.38 |
| 155,172.85 | 145 | 75 | 935,367 | 155,172.85 |
| 206,096.27 | 267 | 99 | 1,230,641 | 206,096.27 |
| 252,280.43 | 275 | 123 | 1,519,752 | 252,280.43 |
| 306,653.64 | 388 | 149 | 1,843,716 | 306,653.64 |
| 352,615.63 | 920 | 173 | 2,136,066 | 352,615.63 |
Large Transaction
Binary log operations such as sorting and merging behave differently based on whether transaction data fits in memory:
-
Data volume < 10 GB: All binary log operations occur in memory. BPS reaches 500 MB/s.
-
Data volume > 10 GB: Binary log operations spill to disk, reducing BPS to 350 MB/s.
| Transaction size | DT (ms) | BPS (MB/s) | EPS |
|---|---|---|---|
| 0.5 GB | 2,000 | 500 | 24 |
| 1 GB | 4,800 | 500 | 24 |
| 2 GB | 7,800 | 500 | 24 |
| 3 GB | 9,400 | 500 | 24 |
| 4 GB | 12,500 | 500 | 24 |
| 5 GB | 17,000 | 500 | 24 |
| 10 GB | 55,000 | 350 | 22 |
| 20 GB | 115,000 | 350 | 22 |
