Before we dive into the technical details, let's consider a nontechnical solution. I still find that most IO problems can be resolved by balancing the IO workload. This may sound simplistic, but while volume management software does a fantastic job balancing IO within its realm, and RAID arrays do much of the same, when multiple realms exist, they may not be balanced. Let me explain this by telling a story.

A few years back, I was performing a remote firefighting consulting engagement. This analysis was fairly straightforward, and there was an obvious IO bottleneck. It wasn't that the IO subsystem was poorly configured, the IO administrator was not doing a good job, or the SQL was not tuned. It was simply that Oracle's IO requirements were clearly exceeding the currently configured IO subsystem's capability.

To complete the consulting engagement, I scheduled a conference call with my client. I asked that not only the DBAs attend, but also that the IO administrator be present. I walked everyone through the OraPub 3-circle analysis and ORTA. Everything clearly pointed toward an IO bottleneck. But then I added that there was some good news related to my analysis. I noticed that while some of the devices had excessive response times and were "hot devices," there were also a number of devices that were being underutilized. In fact, if some of the Oracle database files were moved onto these devices, the IO problem may very simply be resolved. As you might expect, the DBAs were very pleased about this.