Same here. Did some math lately. It is not economically viable for me to accept more than 300 TB of uploads per year and get paid less than ~400 USD per year due to just wear of HDDs and SSD caches on disk writes—I’m not even taking bandwidth into account here. This translates at full utilization to an expectaction that on average (TTL or not) pieces will live for 27 days. The 30 days TTL that the current tests use are pretty close, but it’s still not the majority of data my nodes are still storing, so it’s probably fine for now. But I’m watching closely.
These numbers could be improved quite a lot at the cost of non-trivial engineering effort.
The protocol accepts TTL with subsecond precision, and there’s nothing in the node code that wouldn’t accept a TTL of, let say, 1 second. Even better: a node will gracefully accept an already expired piece. No idea about the satellite though.