Servers handle tasks that require more processing power than the average desktop or laptop can supply, applications that need to be always available, and centralized storage and access to files and applications. Depending on the size of your business you may have only one server or you may have multiple servers. In this section we won’t focus so much on the different roles your server can have but rather the underlying hardware that runs the server. We’ll talk about what you want and need in a server and how it varies depending on the role you intend to utilize the server for. We’ll discuss more about the different roles a server can have in the Software Elements section.
Rack or Tower?
There are two basic choices everyone has to make when they choose a server – rack or tower. Towers are generally used by smaller companies that only have one or two servers while rack mounted servers are used by larger companies that have several servers or storage arrays. A tower server is tall, while a rack server is short and long. Rack servers are made to be quickly accessible and are generally more robust and redundant than their tower server brethren. If you are only a one or two server shop tower servers probably make sense. Once you get beyond this, however, you’ll want to heavily consider purchasing a rack and rack servers. It’ll just make life easier.
Central Processing Unit?
Servers these days generally support at least two processors and many support up to four processors. While there are servers that support more than four processors, these are not generally utilized by mainstream businesses. Each processor is generally a quad core processor – in effect giving you sixteen processors on a four processor machine. When you purchase a server it is good to plan for the future – without spending an arm and a leg. You can usually buy a server with a single quad core processor and then add an additional processor later on, if needed, relatively inexpensively. It is more important to look for servers with the capacity to expand (e.g. additional CPU slots) than those that initially have the best and greatest. The need for more than two processors only occurs in very intensive applications – such as database servers or very high utilization file servers.
Every server you buy should have at least 4 GB of RAM. You don’t want 512 MB chips because they will fill up your RAM slots far too quickly. Remember that 32-bit OS’es cannot support more than 4 GB of RAM, so go with a 64-bit OS. Adding more RAM later is a simple process. Most servers these days have eight or more RAM slots, giving you plenty of room for expansion.
RAID stands for Redundant Array of Inexpensive Disks. There are different levels of RAID.
- RAID 0 provides no redundancy, it simply makes two or more drives look like a single drive, giving more aggregate space. If one disk fails you lose all your data. This is not a recommended configuration unless the data is temporary and expendable.
- RAID 1 is a mirrored arrangement. In this arrangement every piece of information is written to two disks. If one disk fails the other disk still has the information. This allows for good redundancy but also is quite high in cost. For every gigabyte of data you have to buy two times that in hard drive capacity. RAID 1 is quick on reads (since it has two locations to read from) but slow on writes (because it has to write to two locations). Overall it is a decent RAID arrangement and may be used for smaller drives – such as the Operating System drive.
- RAID 5 is one of the more popular RAID configurations, especially when hosting large amounts of data. It uses a n+1 methodology. In simple English that means that it keeps enough data on each of the disks that if one disk fails it can rebuild that disk from the information it has on the remaining disks. This is done by using a mathematical formula (a + b = c). If any element fails (e.g. a, b, or c) one can still deduce the missing value (e.g. a + ? = c, 5 + b = 10, obviously b must be five).
- RAID 6 is essentially RAID 5 except the data is kept in even more locations, allowing two drives to fail before too much information is lost. For example, if you lose two drives in a RAID 5 configuration you can no longer regain your data because you now have something like ? + ? = c (e.g. a + b = 10). With RAID 6 if you lose two drives you will still have enough information to rebuild the original data. RAID 6 is becoming popular as storage arrays get extremely large in size and utilize inexpensive, slower disks.
- RAID 10 or more properly RAID 1 + 0. It uses RAID 1 to create multiple redundant disks, then combines all of those redundant disks into one large drive. One can lose one drive from each RAID 1 without data loss, but if one loses both drives from one of the RAID 1’s one suffers total data loss.
Any server you buy should include internal hardware RAID. While there are software RAID options, hardware RAID is faster and more reliable. In Dell these RAID controllers are known as PERC (PowerEdge Raid Controller) cards, in HP they are known as Smart Array controllers. There are many other companies which also manufacture name-brand and generic RAID controllers.
Any server you buy should not only have a RAID controller but should also utilize it. In smaller businesses this is generally going to be a RAID 1 array. If you have room it is good to have a “hot spare” (a third drive which can be used to fail over to if one of the drives in the RAID fails). It is also possible to create a RAID 5 (across three disks) but even in this instance it is nice to have a hot spare as well.
1U, 2U, 4U?
If you purchase a rack server you will have to deal with the question of U’s. U’s are the standard measurement of servers and measure approximately 1.5″ each U. In general I prefer to avoid 1U servers. While they can be useful, their lack of expandability can make them a somewhat dangerous option. For a few extra dollars one can purchase a 2U server that will be able to handle more CPU’s, RAM, expansion cards, and drives. 4U servers are beasts. They generally aren’t needed for most applications but can prove useful in situations where you need a high-performance database server or other extremely fast server for processing large amounts of complex data. In general I would recommend the 2U server be the base of any business, with 1U and 4U reserved for rare occasions, and possibly avoiding 1U altogether.
- Both Dell and HP offer a solid line of servers. Amongst my favorites are the Dell PowerEdge 2950 (a 2U rack mountable server) and the Dell PowerEdge 6850 (a 4U powerhouse rack mountable). HP has its ProLiant DL380 series which are serious 2U servers at a reasonable price. There are numerous other server providers out there, including many smaller outfits that offer additional price discounts, just remember that you want to pick a provider and stick with them rather than purchasing servers from a wide variety of companies.
- If you purchase a Dell server, I highly recommend adding the DRAC (Dell Remote Access Card) Enterprise card to it. This provides boot level access to your server over an IP network and can replace the need for a traditional KVM.