The current design of ClusterXL demands a latency of no more than 30ms and no more than 2-3% packet loss between cluster interfaces.
Connecting cluster interfaces through several switches can introduce latency, which in turn causes a delay in CCP packets between cluster members.
As a result, CCP packets are not sent/received within pre-defined timeouts on cluster members.
Eventually, an interface will be declared as "failed" and member might go Down.