How to Identify Performance Problem and Bottleneck [message #246965] |
Fri, 22 June 2007 11:19  |
Michel Cadot Messages: 17696 Registered: March 2007 Location: Nanterre, France, http://... |
Senior Member |
|
|
How to Identify Performance Bottleneck in 10g:
[Updated on: Sat, 22 September 2007 09:56]
|
|
|
| How to Identify Performance Problem and Bottleneck [message #247207 is a reply to message #246965 ] |
Mon, 25 June 2007 01:18   |
Frank Naude Messages: 3684 Registered: January 2002 |
Senior Member |
|
|
This following resources may also help:
- On site
- Oracle documentation
- Non Oracle paper
These articles may help you understand some key points:
[Updated on: Fri, 28 December 2007 00:33]
|
|
|
| Performance Monitoring [message #271021 is a reply to message #246965 ] |
Fri, 28 September 2007 14:11  |
Michel Cadot Messages: 17696 Registered: March 2007 Location: Nanterre, France, http://... |
Senior Member |
|
|
The following from Joel Garry explains what you have to do to be able to optimize your performances.
- Read Concepts manual.
- Understand that most performance issues come from application issues.
For example, if some silly SQL reads an entire table to get a few rows, you will likely have a lot of unnecessary I/O that won't fill up the SGA.
- Read the Performance manual.
- Understand the optimizer. It can only use the information it is given.
If the statistics it uses are wrong, nonexistent, or skewed in a manner the optimizer doesn't know about, it can choose a silly plan for accessing the data.
Sometimes a full table scan is not silly.
- Understand what plans are and how to use them to understand 4.
- Understand what statspack can tell you.
- Understand when, how and why to use tracing.
- Understand what waits are and how to evaluate them.
- Read and work through books and articles by Jonathan Lewis, Tom Kyte, and Cary Milsap.
- Understand why rules of thumb can be a bad idea for improving the database of customers.
- Understand that tools based strictly on Oracle can be a bit misleading from a systems standpoint, and systems tools can be misleading from Oracle's viewpoint. Simply knowing you have a lot of reads does not mean anything is wrong, after all, what is a database going to be used for? A proper tuning methodology will figure out what critical bottlenecks are, and what to do about them.
- Understand the basics. For example, if you have sequential write-intensive archive writing interfering with random reads and writes for undo and everything else, thrashing a SAN cache, you probably have a configuration problem. If you
have multiple users accessing data, you need to understand how Oracle handles the issues involved.
- Create clear metrics for performance improvement.
- Read Concepts manual.
- Go to #1 above
|
|
|