Scalability and Clustering
371
Titan SiliconServer
Although the registry is replicated across both Cluster Nodes, there are certain failure scenarios
which could result in the loss of recent configuration changes (this problem is referred to as
amnesia
, a condition where configuration changes are lost because of other failures). The SMU
preserves a copy of the registry, thus ensuring that configuration changes are always replicated.
Data Protection
Titan buffers file system data in
NVRAM
until it is written to disk, to protect it from failures
including power loss. When the Titan SiliconServer is configured as a cluster, each Cluster Node
mirrors the NVRAM of the other node, thus ensuring data consistency in the event of a
hardware failure of a Cluster Node. When a Cluster Node takes over for a failed node, it uses the
contents of the NVRAM mirror to complete all data write transactions that were not yet
committed to disk by the failed node.
Cluster Topology
All cluster elements (i.e. the Cluster Nodes and the QD) are typically connected through the
private management network
. This keeps cluster traffic off the enterprise network, and isolates
it from potential congestion resulting from heavy data access loads.
