Using RCU as the example to motivate GC is interesting. It is essentially transferring the responsibility of freeing from the writer to the last reader, who cannot be determined when compiling. It makes a lot of sense.

But this makes me thinking that, if I want more performance, should I further transfer the freeing from the reader to a dedicated batch process? The readers only update a mark / write into a queue or something. Every once in a while, a batch process collects all garbages and compacts. This way the readers don't have random additional overheads.

> Lies people believe about memory management

> The programmer knows the best times to pause for memory management.

In my experience, there are many programs in which the programmer does know the best times to pause for memory management. For example, in games and crypto trading programs, I want to classify all computations into two priorities. I need to do an update / render a frame / compute a trading action in every time period. If it finishes before the deadline, I have nothing to do now and would like to collect some garbages. If the high-priority thing is using all the available time, for example, when the market is very active, I don't care too much about collecting garbages and would like to defer everything that is not strictly necessary to as late as possible.