Week 12 (22)

Methods for managing deadlock

timeout : set max time limit for each tx
waits-for graph : records T_j waiting on lock held by T_k
- prevent deadlock by checking for new cycle ⇒ abort T_i
- detect deadlock by periodic check for cycles ⇒ abort T_i
timestamps : use tx start times as basis for priority
- scenario: T_j tries to get lock held by T_k ...
- wait-die: if T_j < T_k, then T_j waits, else T_j rolls back
- wound-wait: if T_j < T_k, then T_k rolls back, else T_j waits