makssent commented on issue #36972: URL: https://github.com/apache/shardingsphere/issues/36972#issuecomment-3591643753
Hi @strongduanmu, I’ve done everything you asked for and ran **3 tests of 30 seconds each — before and after the changes**. --- ## 📊 Table (30 seconds, *before changes*) | Threads | QPS #1 | QPS #2 | QPS #3 | **QPS avg** | **p95 avg (ms)** | **Queries total** | **Queries avg** | |--------:|---------:|---------:|---------:|------------:|------------------:|-------------------:|-----------------:| | **120** | 13401.83 | 19315.53 | 19254.07 | **17323.81** | **12.18** | **1 559 143** | **519 714.33** | | **160** | 19177.73 | 19550.43 | 19290.10 | **19339.42** | **18.69** | **1 740 548** | **580 182.67** | | **180** | 19379.13 | 19158.87 | 19159.23 | **19232.41** | **21.67** | **1 730 917** | **576 972.33** | | **200** | 18433.47 | 19394.53 | 18912.33 | **18913.44** | **24.79** | **1 702 210** | **567 403.33** | ### Old flame graphs (2 runs) 1: <img width="2000" height="1235" src="https://github.com/user-attachments/assets/0b915191-69fa-49a9-bb17-eacc304d0d14"; /> 2: <img width="2000" height="1235" src="https://github.com/user-attachments/assets/8b76ffcf-5d0a-4269-8f72-e095d9a6a6d7"; /> --- ## 📊 Table (30 seconds, *after changes*) | Threads | QPS #1 | QPS #2 | QPS #3 | **QPS avg** | **p95 avg (ms)** | **Queries total** | **Queries avg** | |--------:|----------:|----------:|----------:|------------:|------------------:|-------------------:|-----------------:| | **120** | 17828.40 | 19665.90 | 19860.93 | **19118.41** | **10.98** | **1 720 657** | **573 552.33** | | **160** | 19912.30 | 19537.03 | 19429.63 | **19626.32** | **18.19** | **1 766 369** | **588 789.67** | | **180** | 20246.13 | 20434.83 | 19990.00 | **20223.66** | **20.84** | **1 820 129** | **606 709.67** | | **200** | 20119.03 | 20309.77 | 20439.80 | **20289.53** | **23.61** | **1 826 058** | **608 686.00** | ### New flame graphs (2 runs) 1: <img width="2000" height="950" src="https://github.com/user-attachments/assets/2c395b37-04bc-4ac9-9f05-ed02bd6a414b"; /> 2: <img width="2000" height="1064" src="https://github.com/user-attachments/assets/052f0e07-dba4-448f-bfce-513e3f8a5ff0"; /> --- ## 📈 Comparison (before → after) | Threads | QPS avg (before) | QPS avg (after) | Δ QPS | Δ % | |--------:|------------------:|----------------:|------:|-----:| | **120** | 17323.81 | 19118.41 | **+1794.60** | **+10.36%** | | **160** | 19339.42 | 19626.32 | **+286.90** | **+1.48%** | | **180** | 19232.41 | 20223.66 | **+991.25** | **+5.15%** | | **200** | 18913.44 | 20289.53 | **+1376.09** | **+7.28%** | --- As you can see, there is an improvement, but it’s not very large. I understand that many things can still be optimized, but I don’t believe it’s realistic to push the Proxy to 4.5×. Right now, in an ideal (but rare) run we get around 2.5×, in good runs about 2×, and on average something like 1.5–2×. I honestly don’t see how this could be pushed to 4.5× as in your example. If needed, I can also provide full I/O metrics, an archive with all logs, or compile everything into a single detailed report. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
