Tips and Traps August 1999
by Lowell Rapaport
High Availability
Two elements that are always sought on servers are high speed and high availability. High speed is easy to understand. You build a server with the fastest CPU, fastest disk storage, fastest networking and a low-overhead operating system. In document management, however, speed is often secondary to high data availability.
What will insure high availability? Start with hardware redundancy. RAID systems are excellent examples of highly redundant systems. They have redundant power supplies and cooling systems, and are frequently fitted with hot spare drives. In the best RAID systems, all the components are hot swappable. Using RAID as a model, other system components, like servers, can be made with similarly redundant components.
Entire servers can be made redundant through server clustering. In a cluster, several machines are connected together with high-speed network lines. If one server goes down, the other servers in the cluster pick up the slack. Clustering is highly dependent on software that can keep the machines in the cluster synchronized. This is especially important for database-dependent applications like document management. Clustering is especially effective if the servers are kept in separate locations and linked together with Fibre Channel lines.
While high-end Windows NT servers often have redundant power supplies and cooling systems, they don’t usually have advanced microprocessor redundancy. Windows NT servers can be clustered using Microsoft’s Wolfpack software, but Wolfpack clusters are limited to just two machines. For now, the only serious high-availability systems are Unix servers from companies like Sun, SGI, Hewlett Packard and IBM.
|
