Rackable Systems OmniStor 4900F Series, User Manual

Page: 55 / 196

Chapter 2 - Topologies and Operating Modes
A Word about Clustering
41
A Word about Clustering
Minimizing Downtime for Maximum Data Availability
So-called open systems, such as Windows servers, just don’t provide the level of availability
that IS managers are familiar with on mainframes. A partial solution to this problem is
server clustering.
Clusters consist of two or more loosely coupled systems with a shared-disk subsystem and
software that handles failover in the case of a node (host) failure. In most cases,
hardware/software failover is performed automatically and is transparent to users, although
users will experience performance degradation as processing is shifted to another cluster
node. In some cases this failover can occur in a matter of seconds.
High availability of data and applications is by far the most compelling reason to go with
clustering technology. For example, the accepted rule is that stand-alone UNIX systems
can provide 99.5% uptime. Adding a RAID subsystem can increase the uptime to 99.9%.
The goal of clustering is 99.99% availability.
Beyond clustering, fault-tolerant systems can provide 99.9999% uptime. At the high end,
continuous-processing systems offer virtually 100% uptime.
Although the increase from 99.5% to 99.99% availability may seem insignificantly small, it
adds up in terms of minutes per year of downtime. For example, assuming a 7x24
operation, 99.5% uptime translates into 2,628 minutes — or more than 43 hours of
downtime per year. In contrast, 99.99% uptime translates into less than one hour (52
minutes) of downtime per year.
Availability figures relate primarily to unplanned downtime. But the advantages of clusters
in terms of planned or scheduled downtime are even more significant. If you figure two to
sixteen hours per month for a server in a large shop.
Planned downtime requires shutting down stand-alone systems entirely. Result: 100% loss
of processing for the duration of the downtime. But, with cluster, you can shut down one
node and off-load the processing to other nodes in the cluster with no interruption of
processing.
High availability is not the only benefit of clustering. In some cases, users may see
advantages in the areas of performance, scalability, and manageability. In reality, you can
expect a 1.6x (80% efficiency) to 1.8x (90% efficiency) performance increase as you go