# Topology Aware Scheduling (`location-labels`)
+ **Theory:** Machine faults are not completely random independent events. 
  + For example, if two servers are deployed in the same rack, the probability of them losing power at the same time will be greatly increased
+ **Implementation:** According to the topology, `TiKV` servers can be labeled with different labels (`idc`, `zone`, `rack`, `host`), and `PD` makes replicas distributed as scattered as possible
+ **Result:** Capable of toleranting `IDC/zone/rack/host` failures

------------------------------------------------------------