Suppose an Oracle response-time analysis, introduced in the "Oracle Response-Time Analysis" section later in this chapter, shows Oracle processes are primarily waiting for multiple IO block requests. I like to call this physical IO, but really the blocks are simply not in Oracle's buffer cache and reside somewhere else. The terms block reads, block requests, block gets, and disk reads are also commonly used. On the other hand, a buffer get is an attempt to retrieve a block buffered in the Oracle buffer cache. This is also commonly called logical IO, because it is not a physical IO request. It is helpful to think of blocks as residing outside Oracle's buffer cache and on disk, and to think of buffers as residing inside the Oracle buffer cache. The Oracle response-time analysis shows these IO requests take an average of 22 milliseconds (ms) to complete, and the Oracle server and background processes are consuming about 30% of the available operating system CPU.

Based solely on this information, you would expect to find a few SQL statements dominating block read consumption. Looking at the SQL statements, sorted by block reads, you see that two statements clearly are consuming the majority of block read requests. The connection between the Oracle and the application subsystems has been made.

Notice that you have not pointed a finger at anyone; therefore, it's unlikely that you have offended members of any group and caused them to take an immediate defensive posture. There is a very high likelihood that performance-improving changes can be made to both the SQL and Oracle. So right away, two teams can get involved in solving the problem!