Here is a performance comparison of NCMP running with 20,000 and 50,000 CM-handles.
Test environment: Nordix jenkins pipeline (https://jenkins.nordix.org/job/onap-cps-k6-performance-tests-plots/K6_20Performance_20Review/)
On January 24th 2025, the following patch was merged to increase of 20K to 50K handles in k6 tests:
140019: [k6] 50000 CM-handles | https://gerrit.onap.org/r/c/cps/+/140019
Tests were run for 1 week. To compile the report, 15 samples were taken while using 20K, and 15 samples taken using 50K. The raw data is in the attached spreadsheet.
| View file | ||
|---|---|---|
|
Here is a summary of the finding - note the Change % below answers the question: how much slower is it than expected?
Test cases | Number of CM-handles | Slowdown | Change | |||
# | Test Name | 20k median | 50k median | Expected | Actual | +% |
1 | Registration of CM-handles | 205 s | 463 s | 2.5 | 2.3 | -10% |
2 | De-registration of CM-handles | 144 s | 488 s | 2.5 | 3.4 | 35% |
3a | CM-handle ID search with No filter | 0.302 s | 0.967 s | 2.5 | 3.2 | 28% |
3b | CM-handle ID search with Module filter | 0.106 s | 0.305 s | 2.5 | 2.9 | 15% |
3c | CM-handle ID search with Property filter | 0.65 s | 2.10 s | 2.5 | 3.2 | 29% |
3d | CM-handle ID search with Cps Path filter | 0.66 s | 2.19 s | 2.5 | 3.3 | 32% |
3e | CM-handle ID search with Trust Level filter | 2.98 s | 10.86 s | 2.5 | 3.6 | 46% |
4a | CM-handle search with No filter | 2.64 s | 8.09 s | 2.5 | 3.1 | 23% |
4b | CM-handle search with Module filter | 3.58 s | 10.20 s | 2.5 | 2.9 | 14% |
4c | CM-handle search with Property filter | 4.07 s | 11.99 s | 2.5 | 2.9 | 18% |
4d | CM-handle search with Cps Path filter | 4.16 s | 12.45 s | 2.5 | 3.0 | 20% |
4e | CM-handle search with Trust Level filter | 6.7 s | 21.5 s | 2.5 | 3.2 | 29% |
5a | NCMP overhead for pass-through read | 16 ms | 34 ms | 1.0 | 2.1 | 108% |
5b | NCMP overhead for pass-through read with alternate id | 40 ms | 90 ms | 1.0 | 2.2 | 124% |
6a | NCMP overhead for pass-through write | 20 ms | 39 ms | 1.0 | 1.9 | 93% |
6b | NCMP overhead for pass-through write with alternate id | 40 ms | 90 ms | 1.0 | 2.3 | 126% |
7 | Legacy batch read operation (events/second) | 280 | 116 | 1.0 | 2.4 | 141% |