It seems like all the hash implementations I've seen have an insertion rate of ~Million/sec. I wonder if it's possible to get at least an order of magnitude faster (single threaded). Are we close to the input string parsing at this rate? Would sorting/caching help a lot?