The following is the minimum networking infrastructure needed:
The same or very similar physical network infrastructure on individual sites is advised for ease of management.
Use of multiple hubs, routers, and switches is advised to avoid a Single Point Of Failure.
Use a sophisticated routing protocol such as OSPF to adequately manage the routing of packets between sites.
Multiple routes to and from each site are advised.
Independent links should traverse separate physical infrastructures .
FDDI requires two DWDM converts for a single FDDI subnet. Multiple FDDI subnets will require 2x DWDM converters.
Firewalls need to allow all Serviceguard related ports through any port-filtering technology; otherwise , the Serviceguard daemons will not be able to communicate with each other.
Arbitrator nodes are used to provide cluster lock functionality. In the event of a catastrophic network failure between sites, the nodes that will form a cluster quorum will be the nodes that can remain in contact with the arbitrator nodes. In this way, they will be in the majority and allowed to form the cluster; the other nodes that cannot remain in contact with the arbitrator nodes will instigate a Transfer Of Control ( TOC ). As such, the arbitrator nodes are simply installed with the Serviceguard software and are members of the cluster. They do not need access to shared devices because they are not required to run any applications; they simply provide the cluster lock functionality as described.
The differences between Metrocluster and Continentalclusters are outlined in Table 28-1 .
Table 28-1. Differences between Metrocluster and Continentalclusters
A single cluster of up to 16 nodes.
Two independent clusters containing up to 16 nodes each and configured as a simple Serviceguard cluster, an Extended cluster, or even a Metrocluster.
Distances are up to 100km, utilizing DWDM.
Distances only limited by the WAN technology implemented.
All data center machines are connected to shared storage devices.
Machines in separate data centers can be connected to their own storage devices; however, you must have some form of data replication between the sites in order for the data between the sites is as up to date as possible.
Networking is via a single IP subnet(s).
Networking is via separate IP networks.
Failover to an adoptive node is automatic.
Failover to an adoptive cluster is not automatic.
In a Continentalclusters solution, the concept of mutual recovery refers to the ability for the "direction of recovery" to be bi-directional , i.e., site A can recover site B and site B can recover site A. With some physical replication solutions, namely HP XP CA /Extended, additional links between remote disc arrays is necessary. This can significantly increase the cost of the overall solution. Alternately, you can decrease the number of redundant links used, but this compromises the overall performance and threatens the availability of the solution.
Continentalclusters normally involve distances in excess of 100km. The propagation delay of sending data beyond this distance means that asynchronous data replication is normally favored. Suing synchronous data replication could have a significant impact on the online performance of applications, because any IO to the "primary" site will not be acknowledged until the write completes on the "secondary" site.