Common Methods And Tool Recommendations For Performance Testing On Tencent Cloud Servers In Singapore IDC Computer Rooms

2026-08-09 10:25:59
Current Location： Blog > Singapore server
Singapore Server

1.

Overview: Why it is important to do performance testing on Singapore IDC

· Objective: Verify the stability and throughput capabilities of Tencent Cloud (Singapore region) CVM in Asia-Pacific access scenarios.
·Scope: Including network bandwidth, TPS/RPS, latency, system resources (CPU/memory/IO) and database performance.
· Scenarios: Typical scenarios such as e-commerce flash sales, interface concurrency, and high concurrent downloads of static sites.
· Indicators: Throughput rate (RPS/TPS), average/median/95/99 latency, CPU/memory peak, packet loss and retransmission rate.
· Methodology: Layered testing (network layer -> system layer -> application layer -> DB layer), first without CDN/protection direct connection, and then add middleware to verify the effect.

2.

Common load generation tools and usage scenarios

· ApacheBench (ab): lightweight HTTP concurrency test, often used to quickly verify RPS. Example: ab -n 50000 -c 200 http://ip/
· wrk: supports multi-threading, suitable for long-term stress testing and Lua script scenarios, example: wrk -t4 -c400 -d60s http://ip/
· siege: Supports URL list and duration stable stress testing, suitable for simulating multi-interface mixed loads.
· JMeter: GUI/non-GUI mode, supports complex business scripts, assertions, distributed stress testing and monitoring data collection.
· Gatling: Scala script, suitable for performance regression testing in continuous integration pipelines.

3.

Network and bandwidth testing tools (iperf3, etc.) and CDN/DDoS considerations

· iperf3: Measure pure transport layer bandwidth, bidirectional test. Example: iperf3 -s (server); iperf3 -c server_ip -P8 -t30 (client).
· netperf: used for TCP/UDP throughput and latency, suitable for fine-grained network benchmarks.
· CDN verification: First connect directly to the server as a baseline, then enable Tencent Cloud CDN to observe changes in edge hit rate and back-to-source bandwidth.
· DDoS protection: After turning on Tencent Cloud Protection (Anti-DDoS), please note that the combined traffic cleaning will affect the observation, and traffic mirroring testing needs to be done outside the cleaning node.
· Indicators focus on: bandwidth usage, packet loss rate, RTT, TCP retransmission, connection establishment time (SYN to ACK).

4.

System and database stress testing and monitoring tools

· sysbench: used for CPU, memory, fileio, and OLTP benchmarks. Example: sysbench --test=cpu --cpu-max-prime=20000 run.
· mysqlslap / sysbench oltp: Test MySQL’s TPS and transaction latency under multi-concurrency.
· dstat / sar / iostat: Monitor disk IO, network, CPU load, and context switching. It is recommended to cooperate with stress testing for real-time collection (every 5 seconds).
· top/htop/perf: used to locate hot threads and system call bottlenecks.
· Logs and APM: Collect application logs, NGINX access logs, and use SkyWalking/Prometheus+Grafana for trend analysis.

5.

Real cases and server configuration examples (including test data)

· Background: A regional e-commerce company deployed front-end and API services in Tencent Cloud Singapore IDC, with a target peak concurrent request rate of 10k requests/s.
· Initial CVM configuration (example): 4 vCPU, 8GB RAM, 100GB SSD, 200Mbps bandwidth (priced per bandwidth), Ubuntu 20.04.
· Test steps: First perform a wrk stress test on the API (direct connection), and then return the same traffic to the source via CDN; at the same time, use iperf3 to measure the bandwidth limit.
· Test command example: wrk -t4 -c1000 -d60s http://cvm_ip/api/endpoint; iperf3 -c server_ip -P8 -t30.
·Result display (aggregated value):
Testing tools Throughput (RPS/TPS) p50 delay (ms) p95 delay (ms) CPU peak (%)
wrk (direct connection) 4,200 35 120 78
wrk (enable CDN back-to-origin) 3,900 28 95 62
iperf3 bidirectional Bandwidth: 180 Mbps RTT: 12 ms — Network CPU: 30

6.

Tuning suggestions, processes and common pitfalls

· Tuning order: Network->System parameters->Application (nginx/java/mysql)->Horizontal expansion/CDN cache.
· Common kernel tuning examples: sysctl -w net.core.somaxconn=10240; sysctl -w net.ipv4.tcp_tw_reuse=1.
· Nginx tuning reference: worker_processes auto; worker_connections 65536; keepalive_timeout 15.
· Database optimization: increasing the connection pool, optimizing slow query indexes, sharding databases and tables or separating reading and writing, and using Tencent Cloud DBCDB/CMysql to enhance stability.
· Common misunderstandings: 1) Fully loaded single-machine CPU does not mean that the system cannot be expanded; 2) Low CDN hit rate will mislead the judgment of back-to-origin performance; 3) DDoS cleaning accidentally damages normal traffic and requires communication with operation and maintenance.

7.

Conclusion and implementation suggestions

· It is recommended to normalize: include the stress test script into CI, conduct regression stress test and save baseline data before release.
·Monitoring alarms: Combine with Prometheus/Grafana to set CPU/RTT/error rate and other threshold alarms.
· Disaster recovery and elasticity: Use Tencent Cloud CLB+Auto Scaling and multi-AZ deployment to improve availability.
·Evidence: All key tests must retain commands, original logs, and monitoring charts as a basis for capacity planning.
· Contact point: If you encounter regional link problems, it is recommended to contact Tencent Cloud technical support and provide iperf3 and packet capture results for easy location.

Tags：Performance Testing Singapore IDC Tencent Cloud CVM Load Testing Ab Wrk JMeter Iperf3 Sysbench CDN DDoS More»
Previous article： Potential Impact On Game Viewing And Fairness In A Competitive Environment
Latest articles
Popular tags
Stability Singapore Network Service Cn2 Deployment Alibaba Cloud Singapore Region Monitoring Log Management Server Health Corporate Website Port Mapping Performance Performance Stability Cheap Packages Instance Specifications Virtual Private Server Technical Evaluation Cost Analysis Vps Rental Server Technology Budget Control Us Vps Shows Singapore Serverless Network Monitoring International Link Selection Dota Auto-chess Performance Evaluation Reasons Loading Speed King Of Glory Alertmanager Overseas Server Automated Monitoring Cloud Service Provider
Related Articles