[scottlamb]

Those techniques sound possible but the path hint approach seems so much simpler to apply than batching splay updates.

Particularly with epoch GC: When there's heavy reading and rare writes, I like using epoch GC so readers never block. Writes then need to create a whole new map. I suppose one could use something like the delayed queue splay tree you're describing, publishing a new map occasionally, and ensuring this doesn't race with other updates. But it'd be a pain to get right when you could just use path hints instead. Is there an overwhelming benefit I'm missing to justify this complexity and the expense of full map rebuilds?