Yeah, I think it was zstd. But the bigger problem was the curve. Doing something O(N*M) where N (file size) and M (writes per second) were both growing was not

Yeah, I think it was zstd. But the bigger problem was the curve. Doing something O(N*M) where N (file size) and M (writes per second) were both growing was not a winning strategy, compression or not.