How does Facebook's F$4$ increase storage efficiency and remain fault tolerant? $($A$)$ Warm blobs are separated from hot blobs for more efficient replication $($B$)$ Two replicas of each volume are on same racks in the same data center, while the third replica is in another datacenter $($C$)$ XOR coding is used to handle datacenter failures and Reed$-$Solomon coding for rack, host and disk failures $($D$)$ F$4$ only needs storage$-$ heavy nodes but no CPU$-$heavy nodes