Now I realize this will not improve relationships with IO vendors, but it does put them on the defensive, and forces them to take a good, long look at what the IO subsystem is really doing!

Figure 4-21. Operating system tracing is an effective way to remove Oracle from the discussion and discover how long the operating system takes to respond to IO calls. In this example, an Oracle-dedicated server process was traced.

Nearly all Linux and Unix systems have an iostat report; Solaris may offer the sar -d report. If you are using a filer, for example from NetApp, by issuing a remote shell sysstat command directly to the filer, you can gather IO statistics similar to those reported by iostat. If you're a DBA, you may not have permission to do this, so you will need to rely on the wait interface and operating system tracing to build a strong IO subsystem case.