Sun's Cluster and the non-clustered IPMP use simple ICMP echo between the active and back-up interfaces to evaluate their health and decide when it's appropriate to fail over.

In fact I believe they send an ICMP echo to the appropriate network broadcast address, so that, as long as the switch, a router or a client machine responds failover doesn't take place, and as soon as the network goes suspiciously quiet a failover is undertaken whether the ethernet link is still detected or not.

As has been said, it's a fairly basic precaution in a clustered environment to trust as little as possible and to examine the actual service you care about as closely as possible.

I also don't believe this is patented, or rocket science.

On Thu 9 Aug, 2001, Sam Cramer cramer@netapp.com wrote:

...

Does anyone have any experience with this? Can someone confirm or deny the failover in the event of loss of dataflow, but not loss of link?

The vif code detects the loss of link status, not the loss of dataflow. Only the loss of link status will cause a failover.

Sam

-- End of excerpt from Sam Cramer