Titan High Availability Clusters
System Administration Manual
370
Titan High Availability Clusters
The following section details how Titan SiliconServers can be formed as clusters to expand their
functionality.
Clustering Concepts
Titan clustering provides the following functions:
•
Nodes in a Titan cluster can simultaneously host multiple EVS, allowing both servers to
be
active
at the same time, each providing file services to clients.
•
The cluster monitors the health of each server through redundant channels. Should one
server fail, the other can take over its functions transparently to network clients, so no
loss of service will result from the failure.
•
The cluster provides a cluster-wide replicated registry, containing configuration for both
servers in the cluster.
Cluster Nodes
Each Titan SiliconServer that is a member of a cluster is referred to as a Cluster Node. In a
cluster, EVS can be hosted simultaneously on both Cluster Nodes. Titan clustering keeps file
services separate from the Cluster Node on which these services reside. Network users use IP
addresses that are associated with the EVS rather than with the Cluster Nodes. This allows for
seamless, automatic failover, or EVS migration, from one Cluster Node to another.
The Quorum Device (QD)
A Quorum Device (QD) allows the cluster to survive a node-to-node cluster communication
failure and preserves a copy of the
registry
, which contains the cluster’s configuration. The QD
resides and runs on the
System Management Unit
(SMU). The SMU can provide QD services for
up to eight clusters in a Server Farm. Refer to
Managing the Quorum Device
for more
information.
Titan clustering makes use of a
Majority Voting Quorum
scheme, which is augmented by the QD.
The majority voting quorum ensures that only one server can access a file system, thus
preserving data integrity. Each Cluster Node and the QD act as voting nodes in the majority
voting quorum scheme. Only when a Cluster Node has quorum will it be allowed access to a file
system. Normally, the two active nodes in the cluster can constitute a quorum and maintain the
integrity of the file systems. However, under certain failure scenarios, both Cluster Nodes will
attempt to use the same storage (this is typically referred to as a
network partition
, a condition
in which each Cluster Node has lost communication with the other node). In this case, the QD
will “vote” for one of the servers, allowing it to maintain quorum and thus granting it exclusive
access to the storage. If there is no partition between servers, the QD does not participate in any
elections.
