If downtime is not an option, and the Web never closes for business, how do you keep your company's doors open 24/7? The answer lies in high-availability (HA) systems that approach 100 percent uptime.

The principles of high availability define a level of backup and recovery. Until recently, high availability simply meant hardware or software recovery via RAID (Redundant Array of Independent Disks). RAID addressed the need for fault tolerance in data but didn't solve the problem of a complete DBMS failure.

For even more uptime, database administrators are turning to clustering as the best way to achieve high availability. Recent moves by Oracle, with its Real Application Cluster, and Microsoft, with MCS (Microsoft Cluster Service) have made multinode clusters for HA in production environments mainstream.

In a high-availability setup, a cluster functions by associating servers that have the ability to share a disk group. As illustrated here, each node has fail-over node within its cluster. If a failure occurs in Node 1, Node 2 picks up the slack by assuming the resources and the unique logic and transaction functions of the failed DBMS.

Clustering can have the added benefit of not being bound by node colocation. Fiber-optic connections, which can be cabled for miles between the nodes in a cluster, ensure continued operation even in the face of a complete meltdown of your primary system.

When a hot-standby model is in place, downtimes may be less than a minute. This is especially important if your service-level agreement requires higher than 99.9 percent uptime, which translates to only 8.7 hours of downtime per year.

Clustering technologies are pricey, however. The enterprise software and hardware must be uniform and compatible with the clustering technology to work properly. There's also the associated overhead in the design and maintenance of redundant systems.

One cost-effective solution is log shipping, in which a database can synchronize physically distinct databases by sending transactions logs from one server to another. In the event of a failure, the logs can be used to reinstate the settings up to the point of the failure. Other methods include snapshot databases and replication technologies such as Sybase's Replication Server, which has been around for decades.

High-availability add-ons to databases are useful but should be understood in the context of a complete HA methodology. This requires a concerted effort toward standardization on each of your mission-critical infrastructures. Fault-tolerant application design with hands-off exception handling, self-healing and redundant networks, and a stable operating system are all prerequisites for high availability.

When you adhere to these standards, enforceable, database-specific HA technologies are sure to lead your enterprise on the path to minimum downtime.