Quote:
You wouldn't want a simple network failure to cause both machines to think the other is down and cause them both to start writing. Maybe folks have implemented a check on the SCSI level these days.


In real world implementations you have dedicated network connections between the nodes that use special protocols, that are both high-bandwidth and low-latency. Back in the 10/100 days this was frequently SCI (scalable coherent interface) which was more like token ring than ethernet. Now, gigE is generally sufficent, though not running TCP/IP.

A best practices Veritas Cluster Server install will require two dedicated network interfaces running GAB/LLT plus a third (the normal network NIC) running TCP/IP. Node-to-node chatter including the heartbeat is sent over all three interfaces -- two of them dedicated to intra-cluster comms via x-over cables, and a lighter weight backup over your normal network connection and switch.

--Nathan


Edited by Mataglap (22/10/2005 00:36)