Enter our December 2007 Giveaway! We're giving away 14 prizes worth several hundred bucks! Enter HERE.
Statistics
Visitors: 925295
| Introduction To Server Technology & Concepts |
|
|
|
| Written by Brent Earls | |
| Wednesday, 11 July 2007 | |
|
Page 1 of 3
Perhaps the single largest difference between servers and desktops is in priorities. For power users the focus of their desktop is almost always on performance. More RAM, RAID 0 hard drive arrays, high end graphics accelerators, and overclocking are all attempts to try to push the envelope and get the most performance possible. For a server however, the key above anything else is reliability; the speed of the server doesn’t matter if it’s only up for an hour a day. To this end almost everything in a server is backed up, redundant or simply easy to fix in the event of an issue. Another noteworthy detail of servers when compared to desktops is that servers still live in the land of proprietary. While most components on desktops are standardized, servers vary greatly depending on the manufacturer. The only truly standard components on most all servers are Hard Drives, CD-ROM, Floppy, Processors and RAM. The layout and form factor of the motherboard, the output and size of the power supply, as well as the case itself are all proprietary based on your company. The pictures for this article are of a Dell PowerEdge 2650. Let's cover the following topics:
Most servers above basic entry level will have room for at least two processors with larger servers supporting four. Above four processors moves the system into the realm of specialty servers and mainframes in most cases. The most obvious reason for multiple processors is increased performance. This is why before dual core/hyperthreading was available on processors, many high end workstations bought entry level server motherboards. Another quoted reason for two processors is redundancy. If one processor goes bad there's a second. While it is possible to configure a server with a second processor and keep it turned off for emergencies, this is almost never done in practice. The server still has to go down in order to turn on the second processor and besides, how many people have ever had a confirmed failed processor under normal circumstances (no, overclocking and -30 degree temperatures don’t count)? The added RAM slots are for the increased memory demands of a server. For most modern servers I/O is the single most important aspect and large amounts of RAM help to support the transportation of massive amounts of data throughout the computer. The PCI slots on the motherboard are what are referred to as PCI-X slots. These slots have been the staple of the server market for many years. The current versions can provide transfer speeds up to ~4 GB/s. This allows for plenty of headroom for gigabit Ethernet connections, SCSI controller cards and fiber channel cards. The power requirements of a server motherboard with two or more processors are massive. You’ll notice the rear of the power supply interfaces directly with the motherboard. This allows the motherboard to accurately monitor the health of the PSU and allows for the motherboard to control the flow of power around the system. It also allows it to pull a lot of power off of the power supply without having 20-30 cables hooking the PSU to the motherboard. Most everything else on the board will be familiar if in a slightly different locations and in a slightly different number to desktop motherboards. Processors For the most part processors in servers are the highest end desktop, plus some. With Intel for example, the Xeon series for servers and the Extreme Edition for high end desktops are essentially the same (with the exception of the socket). The server processors will sometimes have more cache than their desktop counterpart (again to help with the large I/O processing of servers), but the basic technology behind most processors is very similar to those of desktop. There are exceptions of course; the Intel Itanium (and Itanium 2) lines are based off of a native 64-bit platform, meaning they don’t have support for most modern programs based off of the standard x86 platform. There are other higher-end processors which do differ from normal desktop processors but these are made for Enterprise level operations and mainframes, which might be covered in a later article. Another major difference in servers is support for multiple processors. Normal processors do not contain the necessary circuitry (or it isn’t enabled as was the case with Athlon XP vs. Athlon MP) to work in tandem with another processor. In a server grade processor, this technology is vital to allow the system to work with two CPUs. It has long been the staple of IT departments to buy a server configured as you see above, with room for a second processor and more RAM. While more RAM will frequently be added to a server, second CPUs are almost never added. This is due to the extreme cost of an individual processor, the fact that by the time someone decides more CPU power is needed the server is at end-of-life. The complications that come from adding a CPU after the OS is already installed are also a major stumbling block for adding a CPU to a functional system. RAM As mentioned before, RAM is very important in a server. Many modern servers rely very heavily on I/O to perform their duty, be it sending and receiving massive amounts of data or crunching extremely large numbers. Having substantial volumes of RAM is essential to effective server operation. More than simply having a lot of RAM however, servers need special buffered-modules. These modules help to increase data integrity when reading to and writing to the RAM by using special error-checking methods. The downside to this is increased latency to the memory modules. As mentioned in the introduction of this article, servers sacrifice performance for reliability. |
|
| Last Updated ( Wednesday, 11 July 2007 ) |
