- Technology Options for Disaster Recovery Solutions
- Quick Checklist for Deployments
- Campus Cluster Maximum Distances
- Campus Cluster Topologies and Components
- Campus Cluster Configurations
- Performance in a Campus Cluster Environment
- Management Aspects of Campus Clusters
- Related Resources
Campus Cluster Configurations
Cluster configurations are determined by the technology used to access the remote storage.
Sun Enterprise Servers and Sun StorEdge A5x00 Systems
Sun Enterprise servers (3500 to 10000 series) traditionally use SOC+ to attach to storage subsystems. These adapters use GBICs to convert signals to and from fiber to other media. The default is to use short wave GBICs with multimode fiber that can span a distance up to 500m using 50.0-μ fiber. The GBICs in these specific host bus adaptors (HBAs) can be replaced by long-wave GBICs thatusing 9-μ single-mode fibercan span a distance of up to 10 km. The same is true for the GBICs in the Sun StorEdge A5x00 subsystems. Using this technique, nodes and storage can be separated by a distance of up to 10 km without additional FC-switches.
Configurations using these technologies are in production today in many campus clusters around the world, based on Sun Cluster 2.2 software.
FIGURE 3 represents a typical campus cluster with Sun Enterprise servers and Sun StorEdge A5x00 systems in a three-site configuration.
FIGURE 3 Three-Site Configuration With Sun Enterprise Servers and Sun StorEdge A5x00
Fibre Channel Switch-Based Configurations
The newer, fabric-capable Fibre Channel HBAs used by Sun volume servers and the new generation of Sun Fire™ servers do not allow for the replacement of the on-board short wave GBICs. Instead, the long distance is achieved by introducing Fiber Channel switches into the configuration. Sun's switches allow for the replacement of the GBICs to long-wave GBICs, so that one can connect two switches via a single-mode fiber over a distance up to 10 km. FIGURE 4 shows a typical configuration.
FIGURE 4 Two-Site Campus Cluster With T3WGs
This topology introduces two storage area networks (SANs). Data is mirrored across SANs. Additional storage may be attached to the FC switches. In that case, special care must be taken when choosing the mirror disks in the remote site. Mirrors must be on different SANs in different sites to avoid introducing single point of failure.
The configuration rules must be adhered to regarding firmware revisions, port configurations, topologies, and special restrictions for cascaded switches. Refer to the web site sun.com/storage/san for more details.
Sun Enterprise Servers With FC Switches
A new SBus Fibre Channel card opens up the possibility for the Sun Enterprise servers to be part of a full fabric. This adapter enables enterprises to connect such a server with this HBA to a Fibre Channel switch that is used to span the distance in a campus cluster configuration.
Campus Clusters Using Wave Division Multiplexers
WDMs allow for multiplexing several storage and network interconnects over a single fiber. This approach eliminates the need to have more than two fibers between the sites. However, to prevent a single point of failure, two WDMs are needed at each site. This requirement adds significantly to the initial infrastructure cost, but saves money later due to the limited number of fibers needed to connect sites (other clusters or standalone systems can share the same WDM-based infrastructure).
FIGURE 5 depicts a campus cluster infrastructure built at two sites using WDMs.
FIGURE 5 Two-Site Configuration Using WDMs
Other Configuration Considerations
IP Addresses and Subnets
High-availability services are generally reached through a unique IP address that is known through a name service to all of the clients. If one site in a campus cluster fails, this IP address must be failed over to another site. Therefore, all nodes in a cluster must be attached to the same public net (for example, the same IP subnet). This restriction is true for a campus cluster. To resolve this issue, all cluster nodes must be in the same broadcast domain.
Remote Access to All Consoles
It is a preferred practice to be able to access the console of any cluster node via the network (for example, using a terminal concentrator). In contrast to Sun Cluster 2.2 software, terminal concentrators are not mandatory in a Sun Cluster 3.0 infrastructure, due to a different failure fencing mechanism.
In case of failures in the production network and in the cluster interconnects, the only way to detect the status of the cluster nodes is either through the management network (if it is still operable) or through accessing the console ports using the terminal concentrator. Having a terminal concentrator at each physical location eliminates the need to inspect the physical systems or attach a terminal or laptop to the console port. If both of these methods fail, manual inspection is the only way to come to a final conclusion about what happened.
Communication Channels Between Sites
In a disaster situation, communication is critical. Planning for reliable communications is an important part of developing an overall campus cluster solution. Optional technologies include:
Email If the email system shares the data center infrastructure, it may not be available or reliable in a disaster.
Telephones Power supplies and phone lines may be coupled with the other parts of the infrastructure that has been damaged or destroyed.
Cellular Phones Although mobile phones are helpful for local disasters, they may be unreliable in larger disasters due to unreachable networks, limited range, and dependency on a complex infrastructure that may have been affected by the disaster.
Voice Radio Many security staffs have voice radios. Similar systems may be implemented for emergency communications.
Network Considerations for Client Access
Client access to a production system is often as vital as the production system itself. Additional "networking" connections such as ISDN or fax lines may be critical for normal business operations. Enterprises need to take care of client connectivity to the alternate remote site when designing their campus cluster configurations.