Because when Redis forks to persist, every write will copy a 2MB huge page, and that's not good. All the memory will end to be copy-on-write-ed: latency spikes, big memory usage, ...