C. Repair ScenariosWe study how we

C. Repair ScenariosWe study how we apply predictive repair in two scenarios,namely scattered repair and hot-standby repair, which specifydifferent destination nodes for storing repaired chunks of anSTF node. Figure 1 illustrates both scenarios.Scattered repair selects existing healthy nodes in the storagecluster to repair and store the repaired chunks of the STF node.Specifically, to repair a chunk of each stripe, scattered repairshould choose the healthy node that currently does not storeany chunk of the same stripe, so that we maintain the samedegree of (node-level) fault tolerance. Thus, the performance ofscattered repair depends on the cluster scale and the distributionof the stripes across the cluster.On the other hand, hot-standby repair deploys one or multiplededicated nodes (called hot-standby nodes) to repair and storethe repaired chunks of the STF node. Before repair, such hotstandby nodes serve as backup nodes without participating innormal applications, but take over the service of the STF nodeafter repair. The performance of hot-standby repair dependson the number of hot-standby nodes available for repair.