The Customer is Always Right
For the sake of argument, however, letıs stick with the premise that youıre building this solution in-house. After deciding how many servers you need, youıll quickly have to address the issue of quality of service, and that means fault tolerance. Any server that is conducting business with the outside world needs to be rock solid and stay that way.
 There are many ways to approach this question, mostly guided by your wallet. On the server side, you can opt for fast backup, fault tolerant servers, server clusters or any combination thereof. Defining fast backup is easy: All data is religiously backed up to tape, CD or an outside disk array so that it can be quickly retrieved in case of an emergency. Frankly, this fail-safe needs to be in place no matter what.
Fault-tolerant servers and clusters are often confused for each other. A fault-tolerant server is a single server chassis that holds one or more servers inside in the form of redundant components. These systems carry two or more of everything: processors, disk arrays, RAM banks, power supplies, etc. If one component fails, its backup smoothly takes its place causing no interruption in service and alerting network managers to replace the defective component.
A server cluster, on the other hand, offers a similar kind of functionality but with two or more boxes instead of one. In effect, clusters create a mini-peer network between themselves and share all data and transactions that way. If one server dies, the other takes over while alerting mission control of the problem. What makes clusters more attractive than fault-tolerant solutions is that they can also employ load-balancing. Based upon application or traffic policies, network managers can design clusters to route processing chores to each otherıs CPUs based on availability, thus ensuring much smoother service for customers in high-volume traffic situations. Remember, however, that load-balancing is not parallel processing. In parallel processing, multiple CPUs work together on the same computations. Load-balancing just distributes separate tasks to idle CPUs.
Deciding between these options is largely a matter of what you can afford weighed against what you really need. Fault-tolerant servers, for example, are still very expensive especially when you move into RISC territory, such as Sun Microsystems Enterprise servers or IBM RS/6000s. Configured in a fault-tolerant state, either of these machine types can easily run into six figures a pop. While the sharp mind will quickly see that clustering four or five Intel-based servers together could bring some major cost savings to the table, that is unfortunately usually not the case at the moment.
Thatıs because most of those Intel boxes will be running Windows NT. While Microsoft has implemented a fail-over module, dubbed Wolfpack, this is presently limited to only two machines. While you can opt for third-party clustering solutions like Vincaıs Stand-By Server, these are often expensive and donıt always provide for load-balancing as well as fail-over. Running other Intel-based operating systems, such as Solaris for Intel, Linux or OS/2 may open more options, but be sure to test such solutions carefully as they are not the norm today and so remain something of a dark horse solution.
|
REPORTS
ANALYTICS
Follow 
