Chapter 16. The VERITAS Cluster Server for AIX 709
Low latency transport (LLT): Low latency transport operates in kernel
space, supporting communication between servers in a cluster, and handles
heartbeat communication. LLT runs directly on top of the DLPI layer in UNIX.
LLT load balances cluster communication over the private network links.
A critical question related to cluster communication is, “What happens when
communication is lost between cluster servers?” VCS uses heartbeats to
determine the health of its peers and requires a minimum of two heartbeat paths,
either private, public, or disk based. With only a single heartbeat path, VCS is
unable to determine the difference between a network failure and a system
failure. The process of handling loss of communication on a single network as
opposed to a multiple network is called
jeopardy. So, if there is a failure on all
communication channels, the action taken depends on what channels have been
lost and the state of the channels prior to the failure. Essentially, VCS will take
action such that only one node has a service group at any one time; in some
instances, disabling failover to avoid possible corruption of data. A full discussion
is included in “Network partitions and split-brain” in Chapter 13, “Troubleshooting
and Recovery”, in the VERITAS Cluster Server User Guide.
16.6 Cluster installation and setup
Installation of VCS on AIX is via installp or SMIT. It should be noted, however,
that if installp is used, then LLT, GAB, and the main.cf file must be configured
manually. Alternatively, the installvcs script can be used to handle the installation
of the required software and initial cluster configuration.
After the VCS software has been installed, configuration is typically done via the
VCS GUI interface. The first step is to carry out careful planning of the desired
high availability environment. There are no specific tools in VCS to help with this
process. Once this has been done, service groups are created and resources are
added to them, including resource dependencies. Resources are chosen from
the bundled agents and enterprise agents or, if there are no existing agents for a
particular resource, a custom agent can be built. After the service groups have
been defined, the cluster definition is automatically synchronized to all cluster
Under VCS, the cluster configuration is stored in ASCII files. The two main files
main.cf and types.cf:
main.cf: Defines the entire cluster
types.cf: Defines the resources
These files are user readable and can be edited in a text editor. A new cluster
can be created based on these files as templates.