[Performance] Optimize memory usage #24
Labels
difficulty-difficult
enhancement
New feature or request
perf
Performance related
priority-normal
tracer
Core tracer component
For now there's lots of clones involved in the code. We could use something like https://github.com/thomcc/arcstr to optimize the memory footprint.
The tui can already handle 100000 events with a very low idle cpu usage without hassle thanks to the complex multi-layer cache system I implemented. But the memory usage doesn't look nice to me (total ram 16GB).
The text was updated successfully, but these errors were encountered: