Compaq ProLiant Clusters HA/F100, Administrator'S Manual, Page: 77 / 207

Designing the Compaq ProLiant Clusters HA/F100 and HA/F200 2-43
Compaq Confidential – Need to Know Required
Writer: Bryan Hicks Project: Compaq ProLiant Clusters HA/F100 and HA/F200 Administrator Guide Comments:
Part Number: 380362-003 File Name: c-ch2 Designing the Compaq ProLiant Clusters HAF100 and HAF200.doc Last Saved On: 8/24/00 12:00 PM
Another example of a direct-connect device is a directly connected mainframe
interface. If the first server is directly connected to the mainframe, as through
an SDLC (Synchronous Data Link Control) card in the server, there is no way
to switch the physical connection to a second server. In a case like this, you
may be able to use the client network to access the mainframe using TCP/IP.
Since TCP/IP addresses can be configured to fail over, you may be able to
reestablish the connection after a switch. However, many mainframe
connectivity applications use the Media Access Control (MAC) address that is
burned into the NIC to communicate with the server. This would cause a
problem because MAC addresses cannot be configured to fail over.
Carefully examine the direct-connect devices on each server to determine
whether you need to provide alternate solutions outside of what the cluster
hardware and software can accomplish. These devices can be considered
single points of failure because the cluster components may not be able to
provide failover capabilities for them.
Manual vs. Automatic Failback
Failback is the act of integrating a failed cluster node back into the cluster.
Specifically, it brings cluster groups and resources back to their preferred
server. the Microsoft clustering software offers automatic and manual failback
options. The automatic failback event will occur whenever the preferred server
is reintegrated into the cluster. If the reintegration occurs during normal
business hours, there may be a slight interruption in service for network clients
during the failback process. If the interruption needs to occur in nonpeak
hours, be sure to set the failback policy to “Allow” and set the “Between
Hours” settings to acceptable values. For full control over when a cluster node
is reintegrated, use manual failback by choosing “Prevent” as the failback
policy.
Many organizations prefer to use manual failback for business-critical clusters.
This prevents applications from automatically failing back to a server that has
failed, automatically rebooted, and automatically rejoined the cluster before
the root cause of the original error has been determined.
These terms are described and illustrated in the Group Failover/Failback
Policy Worksheet provided in the following section.