Contents Index Serializable schedules Improving concurrency at isolation levels 2 and 3

ASA SQL User's Guide
  Using Transactions and Isolation Levels
    Choosing isolation levels

Typical transactions at various isolation levels


Various isolation levels lend themselves to particular types of tasks. Use the information below to help you decide which level is best suited to each particular operation.

Typical level 0 transactions 

Transactions that involve browsing or performing data entry may last several minutes, and read a large number of rows. If isolation level 2 or 3 is used, concurrency can suffer. Isolation level of 0 or 1 is typically used for this kind of transaction.

For example, a decision support application that reads large amounts of information from the database to produce statistical summaries may not be significantly affected if it reads a few rows that are later modified. If high isolation is required for such an application, it may acquire read locks on large amounts of data, not allowing other applications write access to it.

Typical level 1 transactions 

Isolation level 1 is particularly useful in conjunction with cursors, because this combination ensures cursor stability without greatly increasing locking requirements. Adaptive Server Anywhere achieves this benefit through the early release of read locks acquired for the present row of a cursor. These locks must persist until the end of the transaction at either levels two or three in order to guarantee repeatable reads.

For example, a transaction that updates inventory levels through a cursor is particularly suited to this level, because each of the adjustments to inventory levels as items are received and sold would not be lost, yet these frequent adjustments would have minimal impact on other transactions.

Typical level 2 transactions 

At isolation level 2, rows that match your criterion cannot be changed by other transactions. You can thus employ this level when you must read rows more than once and rely that rows contained in your first result set won't change.

Because of the relatively large number of read locks required, you should use this isolation level with care. As with level 3 transactions, careful design of your database and indexes reduce the number of locks acquired and hence can improve the performance of your database significantly.

Typical level 3 transactions 

Isolation level 3 is appropriate for transactions that demand the most in security. The elimination of phantom rows lets you perform multi-step operations on a set of rows without fear that new rows will appear partway through your operations and corrupt the result.

However much integrity it provides, isolation level 3 should be used sparingly on large systems that are required to support a large number of concurrent transactions. Adaptive Server Anywhere places more locks at this level than at any other, raising the likelihood that one transaction will impede the process of many others.


Contents Index Serializable schedules Improving concurrency at isolation levels 2 and 3