Summary of how Windows cluster works

Cluster health checks

  • Between every node on all cluster enabled networks
  • By default a heartbeat is sent every 1 second
  • Node is removed from the cluster if 5 heartbeats are missed
  • Not a ping, but a full “Request – Reply” handshake
  • UDP traffic between cluster nodes
  • Sensitive to latency

Intra-Cluster communication

  • Are over single interface
  • Can failover to another interface if there the route fails
    • Networks without network gateway take priority, as it is considered as cluster-dedicated network
  • Includes Resource failover communication:
    • Resource status/state changes (offline/move/online, etc) must be acknowledged by all nodes to the cluster owner between failover steps
  • Sensitive to latency

Microsoft Failover Cluster Virtual Adapter (aka NetFT Virtual adapter)

  • Created and configured when the Cluster service is installed
  • Has APIPA IP address
  • Hidden device in Device Manager
  • “Keep hands off”  – don’t change anything

Microsoft Failover Cluster Virtual Adapter Performance Filter (aka NetFT Virtual adapter Performance Filter)

  • New in Win2012
  • Binds to the physical interface(s)
  • Captures incoming UDP cluster traffic from the network
  • Routes incoming cluster communication directly to the Cluster Service

Cluster service (ClusSvc.exe)

  • Talks to NetFT adapter only
  • Directly via TCP/3343
  • Using IPV6 only

NetFT Virtual adapter

  • Tunnels cluster traffic via the physical interface(s)
  • Communicates on UDP/3343
  • Using IPv4 or IPv6 depending on the physical interface(s) configuration

NetFT Architecture

NetFT-Architecture

Source:

https://blogs.msdn.microsoft.com/clustering/2012/11/21/tuning-failover-cluster-network-thresholds/