Closed GammaPi closed 1 year ago
The current experiments report non-constant memory bound for different running time. A possible reason is memory usage for mysql itself is not constant. Another reason is Scaler memory usage is not constant. We could use an experiment for mysql without Scaler to prove this.
The time increase leads to ~2.94% memory increase
Changing runtime seems to be not a good way to perform this experiment.
Perf's memory overhead collection looks suspicious. But the calculation method is correct.
Already fixed in previous versions.
Currently, context in thread local storage is not freed upon program exits. This cause us not able to claim constant memory bound.
I think the context can be freed easily in new version because I was able to add a TLS check in FGDS module.