How UCloud’s KuaiJie UDB Beats Self‑Built Databases: Performance Test & Cost Analysis

1 Introduction

Yozan's infrastructure uses UCloud services; many databases run on UCloud RDS (some cloud RDS, some self‑built on physical servers). Recently UCloud launched KuaiJie‑based UDB instances with improved CPU, OS kernel, disk, and network, delivering higher performance. This article compares the performance of KuaiJie UDB and self‑built instances to inform future instance selection.

What is KuaiJie?

KuaiJie is a cloud host built on physical machines with a distributed storage layer, optimized CPU, network, and OS kernel. It offers flagship performance with AMD EPYC or Intel Cascadelake CPUs, up to 1000 Wpps network, 120 W IOPS storage, flexible specs up to 96 cores and 768 GB memory, suitable for web, game, database, analytics, and more.

A notable feature is expandable disk space up to 32 TB, which helps avoid costly migrations when storage runs out.

This test uses the Intel‑CPU KuaiJie model; AMD instruction sets may affect MySQL differently.

2 Environment Preparation

UCloud MySQL version 5.7.25. RDS instance uses the default my.cnf template. Core parameters are listed below.

| binlog_format               | ROW |
| innodb_adaptive_flushing   | ON |
| innodb_adaptive_hash_index | ON |
| innodb_buffer_pool_instances| 8 |
| innodb_buffer_pool_size    | 8589934592 |
| innodb_file_per_table      | ON |
| innodb_flush_log_at_trx_commit | 2 |
| innodb_flush_method        | O_DIRECT |
| innodb_io_capacity         | 15000 |
| innodb_lock_wait_timeout  | 50 |
| innodb_log_buffer_size    | 8388608 |
| innodb_log_file_size       | 1073741824 |
| innodb_log_files_in_group | 2 |
| innodb_max_dirty_pages_pct| 50.000000 |
| innodb_open_files          | 1024 |
| innodb_read_io_threads    | 8 |
| innodb_write_io_threads   | 8 |
| innodb_stats_on_metadata  | OFF |
| innodb_thread_concurrency | 20 |
| max_binlog_cache_size      | 18446744073709547520 |
| max_binlog_size            | 1073741824 |
| max_connect_errors         | 1000000 |
| max_connections            | 2000 |
| max_user_connections       | 0 |
| open_files_limit           | 1000000 |
| sync_binlog                | 1 |
| table_definition_cache     | 464 |
| table_open_cache           | 128 |
| thread_cache_size          | 50 |

Parameters are the UCloud defaults with no special tuning.

Previous benchmark articles omitted test machine specs and network latency; this one includes them.

Benchmark Machine Configuration

10 Gbps bandwidth

CPU details

Network latency to UDB: 0.15‑0.22 ms

Network latency to self‑built UHost: 0.22‑0.30 ms; round‑trip time differences affect results.

Benchmark tool: sysbench 0.5.

Test Scenario

100 tables, each with 100 k rows (total 1 GB), DB memory set to 32 GB.

Benchmark script:

#!/bin/bash
threads="12
24
36
48
60
72
84
96
108"
dt=`date +"%Y%m%d%H%M%S"`
mkdir -p /root/yace_$dt
/root/doDBA -log &
for i in $threads; do
    echo $i
    time=`date +"%Y%m%d%H%M%S"`
    /opt/yz-sysbench/bin/sysbench --test=/opt/yz-sysbench/share/sysbench/oltp.lua --oltp-tables-count=100 --oltp-table-size=100000 --mysql-db=sysbench --mysql-user=sysbench --mysql-password=sysbench --mysql-host=xx.xx.xx.xx --mysql-port=3306  --max-time=500 --max-requests=0 --oltp-test-mode=complex   --num-threads=$i run > /root/yace_$dt/thread_${i}_$time.log
    sleep 60
done
p=`pidof doDBA`
kill -9 $p

3 Test Results

QPS

Insert/Delete

Update

Overall, the KuaiJie UDB delivers more than 20 % higher QPS than the self‑built database.

IO‑type benchmark results fluctuated due to short test duration and cloud‑disk jitter; after a firmware upgrade further IO testing is planned.

Price Comparison

UDB cost vs UHost self‑built cost.

UDB: ¥5,160 / month; self‑built on UHost: ¥5,602 / month. The self‑built option also requires additional operational systems (backup, monitoring, HA), increasing labor cost. Therefore, KuaiJie UDB offers better cost‑effectiveness.

4 Conclusion

Recent advances in cloud RDS performance stem from CPU, disk, and network improvements. KuaiJie UDB meets most business performance needs (except heavy bad‑SQL workloads). The test shows a significant performance boost and good price‑performance, making KuaiJie UDB a solid choice for customers.

Have fun with Cloud ^_^