Performance downgraded for larger number of json

[zliucd]

Hi,

There is a major performance bottleneck when using nlohmann:json. In my scenario, there are 300 hundred JSON objects, and each JSON has 1,000 keys. After serializing using json, the performance is downgraded by 8-10x times.

I tried ordered_json and json, and results are identical. I also use Valgrind for performance profiling, and callgrind results indicate json incurs high overhead. As json and ordered_json are built on top of STL, so the real cause is STL?

How to boost the performance? Appreciate your help.

[nlohmann]

Did you compile in Release mode?

[zliucd]

@nlohmann Thanks a lot. I just tested in Release mode, and performance improved significantly(from 8-10x times overhead to 2x), which is a very good result :)