Traditional values

Blade servers may be cutting into the market for traditional servers, but there are still many applications where rack or tower servers are an appropriate and sensible choice.If you have a smaller network, for example, buying one or more individual servers is dramatically less expensive than investing in a blade server. The empty frame for a blade server could cost more than two or three low-end servers.While traditional servers and blade servers perform the same tasks, the main differences are in expandability and processing density.The larger the network, the more rapidly it is likely to grow and the more limited your available physical space is, the more likely you are to want blade servers.Conversely, the smaller the network and the less likely it is to grow quickly, the traditional server is probably a better choice.And as long as you don't begin by installing a maxed-out server, you will be able to expand a tower or rack server's capacity by adding more or larger hard drives, additional memory and supplementary processors.That may not be as simple as adding or swapping blades in a blade chassis, but it is usually highly cost-effective.Where you should never consider traditional servers is in a situation where the initial installation will nearly fill the available space; the servers are fully loaded with memory, storage and processors; and there is the slightest chance that network load will increase.Even if you know that you won't need to expand capacity to serve additional users, your current users will certainly need more processor power as jobs inevitably become ever more technology-intensive and more applications as databases continue to grow in size and complexity.Server-based networks essentially put all your computing eggs in one basket'if the server fails, nothing works.That makes hardware and software dependability critical for any network, usually more critical than mere performance considerations.Without a server you can depend on to be available at all times, the level of performance when it is operating is often a secondary consideration.Getting as close as possible to 100 percent uptime is the goal and, on the hardware side, this affects several parameters.First, you may need redundant storage in the form of a RAID array that automatically duplicates data in case one hard drive fails.Second, the power supply must be capable of not only running the initial processor, storage and memory load, but also any possible expansion in the same chassis.Multiple, hot-swap power supplies are essential for high reliability.Thirdly, cooling capacity is critical but isn't as intuitively obvious as redundant storage and a dependable power supply. Nevertheless, cooling contributes to the reliability of all other components.If you've held a tablet or notebook PC on your lap for an hour or so, you know that computers can be a good alternative to central heating. You just don't notice how much heat they generate in a desktop PC because you aren't in physical contact with the case.[IMGCAP(2)]As a quick and dirty intuitive guide, just think of the power consumption as an equivalent-size bank of light bulbs or space heaters.Now look at those cooling fans'will they really keep all those critical components cool and, therefore, reliable?A server will normally have temperature monitors but, by the time they warn of a problem, the system will probably be close to a hardware shutdown.The final dependability consideration is software-related.While server software is highly reliable, you must also consider how much security the operating system offers and, especially if there is any Internet connection (or worse yet, you have a Web server), you must be extremely careful configuring and maintaining the OS.At a minimum you must disable any and all nonessential services. Just what those are depends on the server software and the jobs the network is called on to do.Some managers mistakenly believe that Linux software is inherently secure'but whether it comes from Microsoft Corp. or some other vendor, all server software must be properly installed and configured'having your network slowed or crashed by a virus or hacker is usually worse than a hardware failure.Also bear in mind that linking multiple servers in a way that one or more can take over the load if another server fails is essential for high reliability. This is often referred to as clustering.Memory, memory, memory'get all the memory you can afford. Once you have a fast processor (relative to the workload) and enough storage, spend every additional dollar in the budget on memory'that will be the limiting factor in determining how happy you are with the server's performance.Buy ample storage to start, but you can always add larger hard drives later and a nearly full drive doesn't degrade performance nearly as much as nearly full memory.A good rule of thumb is that unless you are ruthless in cleaning out old files, your storage requirements will double every year.Consider buying a server with multiple-drive-type support. While it is obvious that you need multidrive support, it isn't as obvious that you may want a server that can support either SCSI or Serial ATA drives.Don't use ATA drives for file storage'they are too slow. But even if your budget lets you get the fastest SCSI hard drives, you may want a SATA drive anyway'as long as it provides the performance you need. The SATA cables are smaller and allow better airflow and therefore better cooling.If downtime isn't critical, you can probably rely on regular backups rather than hardware-based RAID.Make certain you get a server-rated processor such as an Intel Pentium 4 or Athlon MP. Servers put different demands on the CPU than do workstations.Get one or more Gigabit Ethernet ports. They will also run at slower speeds but there is little additional cost and eventually you'll be glad you have them.One final bit of advice: Get a good warranty with a fast service guarantee; this isn't a PC you can swap out. When the server goes down, all of your PCs are useless.