Oracle Server Bottlenecks Oracle Tips by Burleson Consulting

Oracle Server Bottlenecks

If SQL has not been completely optimized, the following server overload conditions are generally true:

I/O overload

This is sometimes evidenced by high db file sequential read and db file scattered read waits and can be detected in the Oracle10g dba_hist_filestatxs view. (SQL that issues unnecessary table block access, possibly due to missing indexes or poor statistics), should be investigated.  Assuming that the SQL is optimized, the only remaining solutions are the addition of RAM for the data buffers or a switch to solid-state disks.

CPU overhead

With the advent of 64-bit Oracle and large data block buffers such as db_cache_size and db_keep_cache_size, the main bottleneck for many databases has shifted from I/O to CPU.  If CPU is listed in the top wait events, sub-optimal SQL that may be causing unnecessary logical I/O against the data buffers should be investigated. The library cache can also be investigated to see if excessive parsing might be causing the CPU consumption.  Assuming that Oracle has been optimized, the options to relieve a CPU bottleneck are to add more CPUs or faster CPU processors. This is an extremely rare event. On most installations which have paid only minimal attention to sizing, CPU bottleneck is suffered only as a consequence of bad SQL. Bad sort or “cached” nested loop will do that. Typically, CPU set is never used more then 25%. Throwing hardware at a problem is typically not a good idea.

Network Overload

In many Oracle-based applications, the largest component of end-user response time is network latency.  Oracle captures important metrics that will show if the Oracle database is network bound, specifically using the SQL*Net statistics from the dba_hist_sysstat view.  Due to the Oracle Transparent Network Substrate (TNS) isolation, there are only a few network tuning options, and most network issues, such as packet sizes, are usually external to the Oracle database.

RAM overload

 As we have noted, the Oracle10g Automatic Memory Management (AMM) utility has facilities for re-sizing the db_cache_size, shared_pool_size, and pga_aggregate_target SGA regions, and Oracle Enterprise Manager for detecting SGA regions that are too small. RAM can be reallocated within these regions which will reduce pga_aggregate_target if there are no disk sorts or hash joins, reduce shared_pool_size if there is no library cache contention, and reduce db_cache_size if there is low disk I/O activity.

Historically, tiny data buffers meant that disk I/O was the most common wait event, but this has changed with the introduction of Solid State RAM disk and 64-bit Oracle where large RAM data buffer caches can be implemented to reduce disk reads. This has shifted many databases from I/O to CPU constraints, and it is one of the reasons that Oracle introduced CPU based costing into the SQL optimizer.

---

This is an excerpt from my latest book "Oracle Tuning: The Definitive Reference".  You can buy it direct from the publisher for 30%-off and get instant access to the code depot of Oracle tuning scripts:

Need an Oracle Health Check? Do you have bad performance after an upgrade?   Need to certify that your database follows best practices? BC Oracle performance gurus can quickly certify every aspect of your Oracle database and provide a complete verification that your database is fully optimized.