3 ms·
Is the next generation of GCs going to use Reinforcement Learning and Machine Learning? The problem with a lot GC is that there's a currently unavoidable tradeo
by ogogmad 4y ago
Is the next generation of GCs going to use Reinforcement Learning and Machine Learning? The problem with a lot GC is that there's a currently unavoidable tradeoff between memory usage, latency and throughput.
- celeritascelery 4y agoHow would machine learning help balance those trade-offs?
- slavboj 4y agoPredicting memory usage patterns can help allocation strategy - eg if you get a flood of traffic every day at 9AM you can time GC to run before the rush, and efficiently preallocate space for the objects you predict needing.