In some other cases, when there is a large number of Static NAT entries on a ClusterXL member, upon failover, the ClusterXL mechanism sends many G-ARP packets. As a result, there is a traffic outage on the network until the routers (or 元 switch) update their ARP cache table. This leads to problems, where the router (or 元 switch) still sends the traffic to the physical MAC address of the old Active member / Pivot member that failed. Reports from the field show that some routers (and 元 switches) do not integrate the information from G-ARP packets fast enough into their ARP cache table.
Upon fail-over in ClusterXL that is configured in High Availability New mode / Load Sharing Unicast mode (Note: any VSX cluster works in High Availability mode), a new Active member / new Pivot member will send a series of Gratuitous ARP Requests (G-ARP packets) for its Virtual IP addresses with the Physical MAC address of the new Active member / new Pivot member on each cluster subnet (and then the new Active member / new Pivot member sends such G-ARP packet every ' N' seconds - the timeout is controlled by the value (in HTUs - HA Time Units) of the global kernel parameter ' fwha_gratuitous_arp_timeout').
0 kommentar(er)
