Choosing the Right Server

You could categorize servers by architecture, type of processor, usage, etc. Talk to any server vendor and you will swim in jargon. RISC or Intel? 32 bit or 64? Rack or blade? IDE or SCSI? You want RAID? Which level? How much redundancy?.... Confused?

Here is a simple overview of how to go about the job of choosing a server. There are no specs given here. For all I know, any specs I give will be outdated by the time you read this. Also, the specs depend on what application you want the server for.

Typically, a server in your organization could fit one of the following models

Work group file and print: This server is used to centrally store data of users and to share things like printers. The load on such servers will not be very heavy and you can use lower-end servers to do the job.

Application servers: These servers run applications such as e-mail, accounting or even things such as ERP. You would not run more than one application on a server. 

Infrastructure servers: This is the machine that runs your firewall, Internet gateway and the like. This one is similar to the workgroup server. There are many organizations that use one or more older PCs to do this job.

The lure of using cheap desktop machines as servers weighs very heavily indeed on the decision maker. Do not. The performance penalties are just not worth the money saved.

Architecture



The first question that you may face is about architecture. Most small and medium enterprises would by default choose the Intel X86 architecture (other choices being the many RISC architectures and the Intel 64 bit architecture).

Processors



Under this architecture you have a choice of processors. Do not use desktop processors for server-class workloads. A server motherboard can take at least two processors. While you can start off with one processor, if you have any plans to add a second processor, it is wiser to do that right at the beginning, as otherwise processor availability and pricing could be an issue.

The choice of the number of processors depends on the specific application you are going to be running on the server

and the number of users that will have to be supported. The standard choices are single processor, dual processor and four way. Of course there are servers that can take more processors, but for the business size that we are considering they would not be required.

Memory



Memory, like number of processors, is based on the application and user load on the system. These days, when 258 MB is becoming standard on the desktop, 512 MB is the bare minimum for a basic workgroup server. For other needs, ask the application vendor for specifications. 

Storage



At one time, SCSI was the only type of hard disk that could be put on a server. No more. Today’s IDE disks are capable of taking entry level server class loads. 

But, choosing server storage is not just a question of choosing the type of disk and the capacity. You could opt for disks on the server, a disk array, separate from the server, or a separate storage device called a NAS (Network attached storage) device.

If you choose direct attached storage (disks on the server), then it is advisable that you opt for a RAID (Redundant Array of Inexpensive Disks) setup rather than having just one or two disks. Simply put, RAID combines multiple disks to perform as one, improving reliability and data throughput in the process.

There are many different types of RAID, known as levels from RAID zero to RAID 5. You would normally not need RAID 5. 

Networking



Servers will have network cards built in. One 100 Mbps card will be standard. Some servers may have a Gigabit card also. If your network is not gigabit, and you do not have immediate plans of upgrading your network to a gigabit one, there is no point in having the gigabit card, and you can actually save some money by deciding not to have it. If your application requires lots of data transfer, then you may get better performance by having more than one network card. The trick is to put half the users on tone card and the other half on to the other.

Expandability



Servers like PCs do offer expandability options. But you are unlikely to make full use of all of them. When the situation comes you are more likely to buy a new server than expand the existing one, unless you are talking about a very high-end server. But basic expandability, like the capability to take more RAM, an extra network card and an extra hard disk should be there.

Chassis, power supply, cooling



This becomes important only if you are buying an assembled server. A server tends to become much more hotter than a normal PC. So there should be enough space inside the chassis for air to circulate and for the heat sinks and fans.

Redundancy



Needless to say, a server is much more critical to your business than any PC. To ensure that the server is available, a high degree of redundancy can be built into it. This starts with a stand buy power supply, all the way to standby RAM And CPU.

In addition, the redundant parts are normally plug and play. That is you can pull out a failed one and replace it, even while the server is running.

When choosing a server for critical applications like ERP, keep an eye on redundancy features.

Form factor



Gone are the days when servers like desktops came in one-size-fits-all configurations. Today you can choose between normal servers, rack servers and blade servers. Blade servers are for organizations that have a very high number of servers and want to fit them into a very small space. So we are not discussing them here. If you are likely to end up with say ten or more servers in one location, it may be worth your while to consider rack servers. These as the name suggests, are servers that are made to look like racks in a cabinet, and actually fit in to a cabinet, making it easier to manage, as well as saving on space.