Closed huntc closed 7 years ago
The following benchmarks show that the disabling vs enabling the new logging shows little overhead. There should also be the potential for higher performance if IO throughput increases.
The results:
setup
sandbox run 2.1.0-beta.1 --no-default-features
test
conduct load visualizer --offline
disabled call
Time 0 | Time 1 | Time 2 | Time 3 | Time 4 | Time 5 | Time 6 | Time 7 | Time 8 | Time 9 | |
---|---|---|---|---|---|---|---|---|---|---|
0m5.342s | 0m3.741s | 0m2.531s | 0m2.411s | 0m2.434s | 0m2.210s | 0m2.156s | 0m2.457s | 0m2.291s | 0m2.152s | 0m2.600s |
new call
Time 0 | Time 1 | Time 2 | Time 3 | Time 4 | Time 5 | Time 6 | Time 7 | Time 8 | Time 9 |
---|---|---|---|---|---|---|---|---|---|
0m4.563s | 0m3.014s | 0m2.461s | 0m2.502s | 0m2.650s | 0m2.259s | 0m2.375s | 0m2.259s | 0m2.500s | 0m2.431s |
Observation tests were also performed for loading bundles from bintray and resolving docker images.
Further performance improvements with all uses of the progress bar considered. The emphasis on the change here is to reduce the amount of floating point arithmetic and perhaps more importantly, the frequency of updates to progress. I think the code is simpler also.
On the update frequency, progress is now updated each tenth of a second.