Chunk buffer size is inconsistent with std recommendation

[HeroicKatora]

NB: stdlib recommends 8k for the size of its own BufWriter.

Originally posted by @nabijaczleweli in #190

The internally used ChunkWriter has a similar purpose. Can we reuse the standard buffer here or refer to the standard definition, or change the buffer size? What are the benefits and motivation of that recommendation?

[nabijaczleweli]

That constant moved there in rust-lang/rust@6d54cd4, and comments reference rust-lang/rust@8128817:

64K was picked for symmetry with libuv, which we no longer use.
64K is way larger than the default size of any other language that I can find. C, C++, and Java default to 8K, and Go defaults to 4K. There have been a variety of issues filed relating to this such as #31885.

The referenced libuv diff (https://github.com/rust-lang/rust/pull/9091/files#diff-b131eeef531ad098b32f49695a031008R62) says

// libuv recommends 64k buffers to maximize throughput
// https://groups.google.com/forum/#!topic/libuv/oQO1HJAIDdA
static DEFAULT_CAPACITY: uint = 64 * 1024;

Relevant mail:

libuv asks for 64K because that lets us (nearly always) drain the kernel receive buffer with a single syscall.
On a tangential note, I've played around with querying the number of pending bytes but the extra syscalls degrade performance by a substantial margin, 25-40% on most benchmarks.

[nabijaczleweli]

I looked at the Writer/ChunkWriter/ChunkOutput implementation and I'm pretty sure one explicitly-buffering sink is required (ChunkWriter), so I don't think that can be replaced with an stdlib BufWriter. Well, could be replaced with having a buffer in Writer itself and implementing Write on that, but that's equivalent.