You have to make good use of preferred and available option on your services and scatter them strategically just like on the image here http://www.facebook.com/photo.php?pid=7118589&l=cd58bfb8e4&id=552113028 that's a 4 node RAC, that simulates the workload of the remaining nodes in the event on of the node goes down. this is also the same concept for SGA provisioning http://www.facebook.com/photo.php?pid=7076816&l=beea222cd0&id=552113028 and cpu core requirement provisioning http://www.facebook.com/photo.php?pid=6973769&l=9b4b053f64&id=552113028 think of it as a massive consolidation exercise where you have to take into consideration the resource consumption on an instance level and also on a cluster level.. http://www.facebook.com/photo.php?pid=7017079&l=72efd9ea41&id=552113028 There are some references here http://goo.gl/ax4dK Also in 11gR2, the QoS was introduced to make use of policy managed databases http://goo.gl/FcCX5 and utilizes server pools to have that "true grid layer" to be able to automatically stand up instances in any available server/host.. that's more of an "automatic" thing which is pretty cool for large clusters. The one that I've mentioned above is managing it "manually". references here QoS Management in a Consolidated Environment OOW 2011 Presentation http://www.oracle.com/technetwork/products/clusterware/qos-management-oow11-1569557.pdf, mixed workload QoS http://www.soug.ch/fileadmin/user_upload/Downloads_public/Breysse_Exadata_Workload_Management.pdf QOS ppt http://www.slideshare.net/prassinos/oracle-quality-of-service-management-meeting-slas-in-a-grid-environment -- Karl Arao karlarao.wordpress.com karlarao.tiddlyspot.com -- //www.freelists.org/webpage/oracle-l