September 1999

REDUNDANCY — Evaluating Your Needs

BY CURTIS NELSON

"How much redundancy do I need"? has quickly become the key question when building or upgrading computer telephony applications. If you are investing in computers for mission-critical applications where downtime translates into lost revenue and disgruntled customers, the answer is: As much as you can get.

The term "redundancy," when referring to industrial computers, means extra or back-up components that automatically kick in when the original component fails. Redundant components allow the computer application to run without interruption in the event that the backed-up component fails.

Further, the term "hot swappable" indicates the redundant components, once failed, can be replaced while the system is operating, thus eliminating any downtime to replace the defective component. Clearly, redundancy and hot-swappability combine to minimize system/application downtime and maximize system quality and service.

HOW MUCH IS ENOUGH?
There are two basic redundancy choices — partial and total. In partially redundant systems only specific components have backup. The most common redundant components are the power supply and hard drive. Together these two components account for 25 to 50 percent of all common computer-related failures, depending on the type of computer, the environment, and the applications.

By investing in computers with redundant power supplies and hard drives, you can protect your system from 25 to 50 percent of all possible failures. However, the system will still be exposed to the other 50 to 75 percent of failure causes. Other common failures include the CPU card, cables and connectors, network cards, voice boards, cache and DRAM, application faults or glitches, and user-induced fault. Downtime to fix such non-redundant components results in lost revenue and service to your customers.

If your goal is to provide for 100 percent fault-free operation, then you need 100 percent redundant components, or total redundancy. Totally redundant computers are the only way to accomplish that goal. For example, if you’re building a network server and you need total back-up, build a completely separate but identical server. If anything fails in one unit, the other unit can take over immediately. When building a communications system that requires multiple computers, plan redundancy into the system. If one computer fails for any reason, a single redundant computer can take over as a spare. Likewise, multiple computers can take over on a load-sharing basis.

Surprisingly, totally redundant computers cost no more than partially redundant computers. With totally redundant computers, you eliminate the cost of redundant components and hot-swappable hardware from a typical industrial computer unit. You also take advantage of all of the components at all times without some components sitting idle, waiting for a failure.

MAKE YOUR SYSTEM HOT-SWAPPABLE
How do you make a 100 percent redundant computer hot swappable? If your application is written to allow your individual computer units to back-up regularly on a network, in the event of a failure the server will sense the failing computer. The server can be programmed to send the backed-up information to redundant/back-up computer(s), which pick-up where the failing unit left off. You can then remove the failed computer for service, re-install it, and bring it back on line without interrupting your application.

This transfer of units takes place while your system is running and therefore by definition is hotswappable. In addition, the ease of removing and servicing single units without interrupting your application is especially significant when upgrading hardware or software.

DON’T PUT ALL YOUR EGGS IN ONE BASKET
Another question to consider when planning for total redundancy is: How much functionality should you install in one computer? The more functionality in a single computer, the greater the computing power needed to run it.

Also, consider how the service would be affected when the unit needs to be removed for service or upgrading. For example, if your application uses 12 voice boards, and you are installing them in a 19-inch industrial chassis, you will need the appropriate computing power to handle the load (in this case, an Intel Pentium 133 with 64 MB DRAM is recommended). On the other hand, you could provide 12 voice boards in the same 19-inch rack space using four separate, smaller units (three voice boards per computer). By doing so, you reduce the actual computing power required per computer (in this case, an Intel DX4/100 with 16 MB DRAM is recommended).

This type of system spreads your application out over multiple computers to minimize the risk of application downtime from a single failure. It also reduces redundancy expense, making service and upgrades easier and far less disruptive to your running application.

If you are using multiple computers for revenue-generating applications, redundancy is critical, added value. Make sure you understand your redundancy needs and evaluate how to best provide redundancy in your system. When downtime equals lost revenue, anything less than 100 percent redundancy is not enough.

Curtis Nelson is president of Crystal Group, a manufacturer of industrial grade fault-tolerant computers specializing in the design and manufacture of space-efficient computer systems for the fast-paced networking and communications market. For more information, please contact Crystal Group at 800-378-1636 or www.crystalpc.com.