Anomalies

Lost Update Problem

• Update of $T_1$ is overwritten by update of $T_2$.
• Update of $T_1$ is essentially lost forever

Dirty Read Problem

• Update on $X$ of $T_1$ fails, and the changes of $X$ should be undone
• However, $T_2$ already used the incorrect results from $T_1$

Locks

read_lock:

• 属于 shared lock
• 允许其他线程 read，但不允许 write

write_lock:

• 属于 exclusive lock
• 不允许 read，也不允许 write

Granularity of Lock

Definition: size of the data item being locked

• e.g. Block / Table / Tuple (record) / Field (column)

• does not impact correctness, does impact performance

  Lock Size:	Small	Big
  Lock conflict prob.	Low	High
  Concurrency	High	Low
  Overhead	High	Low