> No, that's no true. Tracing GC is mainly O(f(live stuff)), not O(f(dead stuff)), so there is basically no problem making lot's of garbage if you don't have an

> No, that's no true. Tracing GC is mainly O(f(live stuff)), not O(f(dead stuff)), so there is basically no problem making lot's of garbage if you don't have any exotic latency requirements. That's true for copying generational garbage collectors. You only need to copy the live stuff and forget everything else. But in e.g. a mark-sweep collector, sweeping dead stuff takes time. In particular, e.g. Chromium's Oilpan can take a long time in the sweeping phase, when there are lot of dead objects. But when there are a lot of microallocations, you also have a lot of live objects, which could otherwise be just a single object in the case of e.g. big arrays. >> Another reason is eliminating unnecessary pointer indirection, which helps make sure that instead of loading data with two MOVs from memory polluting the cache, you need only one. > Uh, that is locality, what I started with, no? Hmm, seems I highlighted the wrong aspect. It's not just locality, but also size of the data you're dealing with. Those pointers take space. Another aspect of it is that arrays of unboxed elements have more predictable access patterns. Iterating over an array of unboxed elements is friendly to cache prefetching. CPUs recognize this (linear) pattern. With boxed elements and a compacting garbage collector locality may be fine, i.e. elements of the array may be close in memory, but the order in which you fetch elements is pretty random. When the cumulative size of the elements of the array is significantly larger than 64 bytes (the usual size of a cache line), then you're going to be prefetching the wrong regions of the array. There are situations where the order is going to be right, because e.g. the elements were created in the same order in which they are in the array, but that will be destroyed after some sorting, filtering or whatever.