|
A service group in VCS consists of an application (e.g. - Progress) along with a Virtual IP address and the associated Network Interface Card, Disk Mount Points, Disk Volumes and Disk Groups. The service group can be failed over from host to host within the local area cluster at one site, independent of any other service groups. One of the benefits of VCS therefore is it is possible to configure multiple service groups in a single cluster and have each service groups only failover to another host when there is an issue with one of the resources required for that service group to run. With multiple physical NICS in a single host, it is possible to automatically restart a service group on a alternate NIC if the primary NIC has failed.
A separate VCS management service group can be created which can contain the virtual IP address for the web interface to manage the VCS cluster. A failover of a service group from one host to another requires shutting down all of the resources within the group, exporting the disk groups containing the disk volumes for the service group, importing the disk groups in to the failover host, and then starting up all resources within the service group. Failover times can vary but typically most of the failover time is due to the length in time to shut down and restart the main application (e.g. – Progress).
VCS can allow up to 32 nodes to be configured in a single cluster. The management interface is the same whether the cluster is AIX, Solaris, HP-UX, Linux, or Windows based. The failover can be N to 1 (where N is up to 32 nodes) or even N to N failover (where no single node acts as a passive standby host).
VCS uses fast dedicated protocols known as Global Atomic Broadcast (GAB) and Low Level Transport (LLT) to provide host to host communications within the cluster. GAB and LLT are much more efficient than using TCP/IP as the cluster communications protocol and this efficiency allows the cluster to scale up to 32 nodes in size. With a two node cluster, GAB and LLT can run over crossovers Ethernet cables where each crossover cable represents a cluster heartbeat link. If more than two nodes are to be run in the cluster, then an Ethernet switch or hub is required to interconnect all of the hosts. It should also be noted that a VCS cluster requires an absolute minimum of two heartbeat links for redundancy and can be configured with up to 8 heartbeats.
Data that must be replicated between two sites can be done at a disk volume level through the licensing of a product known as Symantec Volume Replicator (VVR). VVR can replicate asynchronously (or synchronously if the replications are “close” in terms of LAN distance such that write performance is not adversely affected.) Volumes to be replicated are set up within a replicated volume group (RVG) and the replicated data is sent via TCP/IP to the receiving host(s). Write-order fidelity is guaranteed with VVR.
Global Cluster Manager (GCM) allows control of failover of the service groups from one location to another location. GCM combines with VVR to enable automated failover of the application.
|
