QPS fails to go up due to frequency limitation of public network CLB bandwidth

【 Abstract 】

some KA The production environment of the project is due to the public network clb Frequency limiting results in qps Couldn't get on

【 Problem phenomenon 】

After pressure measurement under different concurrent conditions , Discover that performance data is always 20qps ～ 30qps Section , Unable to rise further

【 Troubleshooting ideas 】

1. Comb the link 


Chengdu piezometric cluster 4 platform -> DNS -> High prevention -> waf/ Public network （clb）-> Access gateway -》 Access gateway -》ngnix -》 Application service -> tdsql database （ Lord / To prepare ）

2. Check whether the link node has a load bottleneck , No bottlenecks found 


3. Shorten the link through troubleshooting , Conduct internal and external network voltage measurement ,qps Still low , It has nothing to do with the external network link 


4. Try to bypass the intranet clb Pressure measurement , Find out qps Very high , Identify bottlenecks and net、clb of 


5. see net found 100M bandwidth , Increase the bandwidth to 1.5g,qps rising 3 times , however qps Still very low 


6. see clb found 100M Bandwidth limitation , Turn it up to 1.5G,qps rising 5 times ,qps Pretty much as expected

Discovery public network clb Bandwidth default 100M, Adjust its bandwidth 1.5G,qps from 20 Rise to 200

Public network CLB Bandwidth default 100M, But during the pressure measurement , Flow on 15M, and qps It can only reach 20s. Adjust the bandwidth limit 100M rising 1.5G,qps from 20 Rise to 200.

100M bandwidth

1.5G bandwidth

The bottleneck of pressure measurement is CLB（ Public network ）, Across clbqps You can achieve 2000+

Public network clb Bandwidth default 100M, Flow on 15M,qps20. Bandwidth limitation 100M Rise to 1.5g,qps from 20 Rise to 200