look here for the node fencing:
Yes RAC can reboot a server as some processes run with root privilege. It does not matter whether this is a virtual of physical server.
The Oracle Clusterware is designed to perform a node eviction by removing one or more nodes from the cluster if some critical problem is detected. A critical problem could be a node not responding via a network heartbeat, a node not responding via a disk heartbeat, a hung or severely degraded machine, OS resource starvation (i.e High cpu usage , Memory shortage/Swapping , High run queue/load average) or a hung ocssd.bin process.
Oracle Clusterware relies on the accessibility of the Voting File(s). If a node in the cluster cannot access the majority of the Voting File(s), the node (s) to which this applies is (are)
immediately removed from the cluster (evicted / fenced).
The purpose of this node eviction is to maintain the overall health of the cluster by removing bad members(in other words, to prevent split-brain situations, as well as data corruption).
Note: Starting in 184.108.40.206 RAC (or if you are on Exadata), a node eviction may not actually reboot the machine. This is called a rebootless restart. In this case we restart most of the clusterware stack to see if that fixes the unhealthy node.
For more info on 11gR2 RAC : REBOOT-LESS NODE FENCING, refer the below URL:
The network heartbeats are associated with a timeout called misscount
crsctl get css misscount
crsctl get css disktimeout
One point of clarification in case it was obvious from the other posts...RAC does not reboot anything. Real Application Clusters is an option within the RDBMS software that allows a clustered database. RAC must be installed on top of some clusterware software. Most people use Oracle's Clusterware (now called Grid Infrastructure). It is the Clusterware software that detects split brains, fencing, and other cluster issues. If clusterware determines it to be the proper course of action, a node will be evicted from the cluster. When a node is evicted from the cluster, the clusterware software running on that node may attempt to reboot the node in the hopes that once it comes back up, it can successfully rejoin the cluster. RAC does not do any of this.