Oracle Core: Essential Internals for DBAs and Developers

How Can It Work?

When you look at a picture of a cluster of eight machines, all running an instance of Oracle against the same database, you’ve got to wonder how one instance can make sure that the block it wants to modify isn’t being modified by the other seven instances at the same time—and won’t the problem become harder as the number of instances increases. This is where the genius of the implementation comes in—once you’ve got to three instances the level of overhead doesn’t get any worse. To see how this works we’ll start by looking at the global resource directory or GRD that is a distributed mechanism for controlling access to “things,” and cache fusion that allows the most important things (the data blocks) to move around the cluster ...