What kind of server do
you need?

by Neil Randall and Allen Fear (3/1/2004; updated 2/28/2005)

What kind of server do you need? Should you get a single CPU or two? One hard drive or two? SCSI or ATA? To RAID or not to RAID? Would a less expensive NAS unit suffice?

It would be nice if there were a mathematical formula that took the number of clients and types of applications as inputs and spit out a specific server configuration that would meet your needs. But there isn't. Rather, there are a series of questions you need to consider before you shop online or call some friendly local resellers and national vendors for their specific buying advice.

How many clients will hit the server?
As noted, the answer to this one won't define the server you need. But it'll establish the parameters of the subsequent discussion. Just to give you an idea, though, according to Tony Woodard, systems engineer at MPC, performance curves on servers using one or two Xeon processors offer pretty similar performance until you hit 20 users, then they start diverging.

What kinds of applications will your server support?
As Jeff Carlat, a marketing guy at HP, puts it, "Fifty people checking their e-mail every once in a while is a very different thing than fifty people streaming video over the network all day long."

How important is your server to your business?
If your business would go down in flames five minutes after your server crashes, reliability and redundancy quickly rise to the top of your must-have list. At most vendors, the higher you go on the product line chart, the more redundancy you get. Managed hosting services are a compelling alternative for businesses who are open to the idea of outsourcing IT resources. In a nutshell, these services allow you to rent computing power and disk space offsite for a monthly fee that includes the management of the technology resources and obviates the need for and expense of building and maintaining an IT infrastructure in-house.

That redundancy can take the form of RAID (Redundant Array of Inexpensive Discs), which mirrors the data from one hard drive onto another, so if the first goes the second can take over. It can also take the form of ECC (Error Checking and Control), which is like RAID for your memory; if a memory module misplaces a bit, the system will keep running, instead of issuing a Blue Screen of Death. Then there are servers with redundant power supplies, even redundant fans.

Note that question number three is actually related to the first two: The more stress you're putting on a system, in terms of users and applications, the more likely it is the system will fail. ("Thermal events"--a consequence of hardware running too hard--are the top cause of server failures, according to HP's Carlat.)

What is your budget?
This is the final and probably deciding factor in your buying decision. While low-end servers are often advertised at extremely low prices, those advertised prices don't get you much in the way of real functionality. In many cases, you'll have to upgrade those base configurations considerably to get a server worth the name. How much you upgrade--and how far up a vendor's product line you can climb--will largely depend on how much money you want to spend.

As Jeff Carlat puts it, "It's like buying a house: You get as much as your budget allows and as much as your expected growth rate requires."

Neil Randall teaches software interface analysis and design at the University of Waterloo, Canada, and has published numerous books and articles covering operating systems, networking, and hardware. Allen Fear is a senior editor at CNET.