RAC uses a shared disk; it doesn't eliminate the single point of failure. The shared disk also adds a contention point, such that RAC often stops scaling after

RAC uses a shared disk; it doesn't eliminate the single point of failure. The shared disk also adds a contention point, such that RAC often stops scaling after a handful of nodes. Even the first few don't give you linear scalability without changes to the application and extensive tuning. I'm not trying to defend the article though.