The modern GC implementations try to minimize blocking "stop-the-world" stalls by doing as much work as possible on the background (i.e. on a separate thread), for example marking unreachable garbage instances while the application process continues to run. In spite of these advancements, for example in the .NET CLR paradigm it is still very difficult to maintain large heaps (millions of objects) with resident objects that get promoted to older generations without incurring noticeable delays (sometimes a few seconds).