The cache TTL on VOD is significantly higher, so your whole infrastructure needs to have a live origin, which gets hit repeatedly to get the latest fragments/segments from the origin, from multiple edges/midgresses. I find that our origin consumes an order of magnitude more CPU than transcoding and packaging for live once you have even a couple of thousand live viewers and when serving LL-HLS/ll-dash with a low fragment size it increases more. if you are latency sensitive then i between caching layers are not an option.