Once the load balancing protocol is initiated, the requesting processor starts searching for candidate processors for task migration.
The selection process can be randomized, static, or adaptive. The randomized scheme picks an arbitrary set of processors; the static scheme select a set of neighbor processors that is determined a priori; the adaptive scheme uses global information collected at runtime to determine what processors to choose.
The randomized scheme sets up direct migration paths between processors. However, it may be inefficient in finding valid candidate processors. The static scheme is extremely simple, but may cause more overhead since tasks may be moved several times before it reaches its destination. The adaptive scheme performs best when the global information is accurate. However, there is a trade-off between the accuracy of global information and the overhead of collecting it.