Oracle Essentials [Electronic resources] : Oracle Database 10g, 3rd Edition نسخه متنی

7.6 Concurrent Access and Performance

When you read through all the steps
involved in the above processes, you may assume that Oracle would be
a very slow database. This is not at all true. Oracle has
consistently turned in benchmarks that make it one of the fastest
databases, if not the fastest, on the market today.

Oracle provides good performance
while implementing multiversion read consistency by minimizing and
deferring unnecessary I/O operations. To assure the integrity of the
data in a database, the database must be able to recover in the event
of a system failure. This means that there must be a way to ensure
that the data in the database accurately reflects the state of the
committed data at the time of the crash. Oracle can do this by
writing changed data to the database whenever a transaction commits.
However, the redo log contains much less information than the entire
data block for the changed data, so it's much
"cheaper" to write to disk. Oracle
writes the redo information to disk as soon as a transaction commits
and defers writing the changed data blocks to the database until
several sets of changed blocks can be written together. Oracle can
restore the database using the redo logs, and these procedures cut
down on time-consuming I/O operations.

However, when you're considering the performance of
a database, you have to think about more than simple I/O operations.
It doesn't really matter how fast your database runs
if your transaction is waiting for another transaction to release a
lock. A faster database may complete the blocking transaction faster,
but your transaction is still at a dead stop until the blocking
transaction completes. Because most databases perform a mixture of
reading and writing, and because Oracle is the only database on the
market that doesn't use read locks, Oracle will
essentially always deliver the lowest amount of database contention.
Less contention equals greater throughput for a mixed application
load.

There is also more than one type of performance. Performance for
database operations is measured in milliseconds; performance for
application developers is measured in months. Because Oracle provides
much less contention with its read consistency model, developers have
to spend less time adding workarounds to their applications to handle
the results of contention.

It's not as though Oracle is the only database to
give you a concurrency solution you can use to implement applications
that provide adequate data integrity. But the multiversion read
consistency model makes it easy for you to get a consistent view of
data without excessive contention and without having to write
workarounds in your application. If it sounds as if
we're big fans of Oracle's locking
scheme, wellwe are.