It is conceptually very simple: the GC stops the mutator threads while compaction is in progress and objects are moved around; throughput-focused GCs generally

It is conceptually very simple: the GC stops the mutator threads while compaction is in progress and objects are moved around; throughput-focused GCs generally do that, better ones do it in parallel with multiple GC threads. This has an obvious impact on performance that needs to be compared with the mutator calling malloc/free in a non-GC system. The function isn't called "free" because it runs in zero time, you know.