Why does v5.27.1 Perform better than v5.26.0 ？

[GitBenjamin]

Hi, @SiarheiFedartsou
I found that when "step=ture", the performance of the v5.27.1 route api was better than that of v5.26.0 I'm curious what have you done to improve performance?

Compare the two versions:

Formula mode: (old_version - new_version) / old_version * 100

route step=false scene 600qps	No change	Reduce by 5%	Mean RT -0.4ms
route step=true scene 400qps	Reduce by 30%	Reduce by 40%	Mean RT -10ms

[danpat]

@GitBenjamin We have a complete CHANGELOG file here:

https://github.com/Project-OSRM/osrm-backend/blob/master/CHANGELOG.md

It shows a summary of all changes merged for each release. You should scan that list and see if anything relevant to what you're testing seems to apply.

My guess would be improved JSON result serialization in #6380 - that would matter when comparing steps=true to steps=false, as there is more JSON to serialize.

[GitBenjamin]

Thanks @danpat, thanks for the tip!
I tried to check the CHANGELOG and now I'm still looking for improvements.

I thought it was caused by the following changes:

• Improvements to maneuver override processing
• Lazily generate optional route path data
• Upgrade Boost dependency to 1.70
• Migrate to C++17. Update sol2 to 3.3.0.
• Fix performance issue after migration to sol2 3.3.0.
• Fix distance calculation consistency.
• Migrate to modern TBB version.
• Pass osm_node_ids by reference in osrm::updater::Updater class.
• Remove redundant nullptr check.
• Improve performance of map matching via getPathDistance optimization.
• Fix inefficient osrm-routed connection handling
• Fix bug with large HTTP requests leading to Bad Request in osrm-routed.

Now I can start with your prompt. I didn't realize that return would also improve performance.

[SiarheiFedartsou]

@GitBenjamin are you using osrm-routed or Node bindings btw?

[GitBenjamin]

@GitBenjamin are you using osrm-routed or Node bindings btw?

I am using osrm-routed.