CLUSTERING 
ERIC VANDERBURG
CLUSTERING 
• Combining computers to create one 
logical computer 
• Provides increased redundancy
TERMS 
• Node (cluster host) - A computer that is a member of a server cluster. 
• Resource - entity that can be managed by a cluster and moved between 
nodes. 
• Resource Group - A collection of resources that can be started or stopped. 
• Failover - the process of taking resources offline on one node and moving 
them to another. 
• Failback - returning resources back to their original node once it has 
recovered. 
• Quorum - the disk that holds the configuration data and all the changes that 
have been made to the cluster database. 
• HBA (Host Bus Adapter) - interface card that connects a cluster node to a SAN. 
• Split brain - some nodes in the cluster lose contact with the quorum and one 
another so each thinks that they are the only one in the cluster. This 
eventually leads to data corruption.
INSTANCES 
• Single Instance application - One application is 
running on the server at a time 
• Multiple instance application - multiple instances 
of the same code or different pieces of code 
running for the same service are running on a 
cluster. 
• Cloned applications - uses a relatively static dataset. 
• Partitioned application - dataset is frequently updated 
so the dataset is divided among partitions.
STORAGE 
• Clusters use SANs for storage 
• Cluster network should be connected directly to 
the SAN and the network between them should 
be isolated from the rest of the network. 
• Parallel SCSI 
• Fiber Channel 
• If there are more than 2 nodes,m FC-SW 
(Switched Fiber Channel Fabric) must be used.
ACCESS TO THE SAN 
• Zoning - creates views called clouds that control 
access and control traffic flow between the SAN. 
• LUN masking - access control is provided at the 
controller and port level. Policies are defined for 
specific devices. Provides more granular security 
than zoning.
CLUSTER MODELS 
• Local Quorum cluster - used for testing and has only one 
node. 
• Single Quorum device cluster - cluster configuration data is 
stored on a single device connected to all nodes in the 
cluster. At least one node must be functional for the 
cluster as a whole to function. 
• Majority node set cluster - each node contains its own 
copy of the cluster configuration data. A majority of the 
nodes in the cluster must function for the cluster as a 
while to function.
MULTISITE CLUSTER REPLICATION 
• Block level replication - storage devices are 
mirrored over a private network between 
sites. 
• File system level replication 
• Application level replication
SYNCHRONOUS & ASYNCHRONOUS 
• Synchronous replication - an operation is 
not complete until it has been replicated to 
partner sites 
• Asynchronous replication - an operation can 
be considered complete before it has been 
replicated to partners.
SYNCHRONOUS & ASYNCHRONOUS 
• Synchronous replication - an operation is 
not complete until it has been replicated to 
partner sites 
• Asynchronous replication - an operation can 
be considered complete before it has been 
replicated to partners.

Clustering - Eric Vanderburg

  • 1.
  • 2.
    CLUSTERING • Combiningcomputers to create one logical computer • Provides increased redundancy
  • 3.
    TERMS • Node(cluster host) - A computer that is a member of a server cluster. • Resource - entity that can be managed by a cluster and moved between nodes. • Resource Group - A collection of resources that can be started or stopped. • Failover - the process of taking resources offline on one node and moving them to another. • Failback - returning resources back to their original node once it has recovered. • Quorum - the disk that holds the configuration data and all the changes that have been made to the cluster database. • HBA (Host Bus Adapter) - interface card that connects a cluster node to a SAN. • Split brain - some nodes in the cluster lose contact with the quorum and one another so each thinks that they are the only one in the cluster. This eventually leads to data corruption.
  • 4.
    INSTANCES • SingleInstance application - One application is running on the server at a time • Multiple instance application - multiple instances of the same code or different pieces of code running for the same service are running on a cluster. • Cloned applications - uses a relatively static dataset. • Partitioned application - dataset is frequently updated so the dataset is divided among partitions.
  • 5.
    STORAGE • Clustersuse SANs for storage • Cluster network should be connected directly to the SAN and the network between them should be isolated from the rest of the network. • Parallel SCSI • Fiber Channel • If there are more than 2 nodes,m FC-SW (Switched Fiber Channel Fabric) must be used.
  • 6.
    ACCESS TO THESAN • Zoning - creates views called clouds that control access and control traffic flow between the SAN. • LUN masking - access control is provided at the controller and port level. Policies are defined for specific devices. Provides more granular security than zoning.
  • 7.
    CLUSTER MODELS •Local Quorum cluster - used for testing and has only one node. • Single Quorum device cluster - cluster configuration data is stored on a single device connected to all nodes in the cluster. At least one node must be functional for the cluster as a whole to function. • Majority node set cluster - each node contains its own copy of the cluster configuration data. A majority of the nodes in the cluster must function for the cluster as a while to function.
  • 8.
    MULTISITE CLUSTER REPLICATION • Block level replication - storage devices are mirrored over a private network between sites. • File system level replication • Application level replication
  • 9.
    SYNCHRONOUS & ASYNCHRONOUS • Synchronous replication - an operation is not complete until it has been replicated to partner sites • Asynchronous replication - an operation can be considered complete before it has been replicated to partners.
  • 10.
    SYNCHRONOUS & ASYNCHRONOUS • Synchronous replication - an operation is not complete until it has been replicated to partner sites • Asynchronous replication - an operation can be considered complete before it has been replicated to partners.