​

​

In the field of servers, processors, memories and hard drives reign, which will surely become part of our desktop computers in the not too distant future; at least, in most cases this has been happening historically. In general, the internal hardware of network servers is not so different from the hardware of a desktop computer. We will see here those subtle differences between both worlds.


Microprocessors

For today's servers, both Intel and AMD offer multi-core processors - up to eight or ten. Opteron is AMD's line of server processors, while Xeon and Itanium are owned by Intel.

Motherboards

Most motherboards allow two, four, eight, and more of these processors to be placed on the same board, thereby multiplying computing power. They also have several sockets to install Fully Buffered RAM, generally with four to eight modules. As for the maximum capacity supported, it varies between 32 and 128 GB.

We should note that these motherboards do not have an integrated audio interface, since it is not necessary. They usually bring a built-in video card, with limited features, because this is not the main section that the servers point to either. In many cases, they integrate a 10/100/1000 Mbps Ethernet network interface.

Regarding the expansion sockets that a typical server-oriented motherboard has, the most common at present is the PCI Express 3.0 16x and the PCI-X, an extension of the classic PCI bus, but which works at 64 bits, and 66 o 133 MHz. The inclusion of an Ultra SCSI 320 disk controller, with both internal and external output, is practically mandatory.

Storage

SCSI 320 and SAS interface disks are the most popular in this area. The rotation speed of these units can be 10,000 revolutions per minute, although there are also models of 15,000 and 20,000 rpm; remember that the disks of a desktop PC rotate at 7200 rpm.

Regarding the capacity of the unit or units used, this depends directly on the tasks assigned to the server and the number of users that must be served, among other factors. The most common is to see drives arranged in such a way that they make up a RAID array, to increase either speed, security, or both.

​

Cabinet

For large servers, the market offers special cabinets, which can be of three types: tower, rackeables or blade. The tower are commonly used on desktop computers, with the difference that they are larger on servers, and they have a large number of bays to accommodate hard disk drives and enough space to accommodate large motherboards. In this case, ventilation is taken into account: they usually have between four and ten built-in fans.

​

Rackable cabinets are modules that can be added and bolted to a box or tower called a rack. Therefore, a rack is nothing more than a set of equipment (servers, switches, routers, patches, etc.) that are stacked in a modular way as if they were shelves.

This method is also used in other areas, such as sound, telephony, communication and medical equipment, since the width of the rack is standard: 19 inches. As for the height, this can be variable. The minimum unit is known as 1U, and there are 2U, 4U, and up to 8U servers. The space of these cabinets is reduced compared to the tower format. Ventilation becomes a priority in this type of server, but being modular, the advantages they offer are broad.

​

Finally, the blade type cases are similar to the rackable ones, with the difference that the first ones are located vertically in relation to the rack and are capable of accommodating up to ten widths of the shelf. This makes better use of the available space in the rack. For example, in a 42U rack, we can typically accommodate 42 1U high devices.

Energy sources

The most widely used and recommended power sources on servers are redundant calls, also known as duals. They allow that, if on one server one of the sources suffers a failure, the other one takes control, while the first one can be replaced, all this, without turning off or restarting the server. Obviously, special motherboards are required for these cases, and the costs are quite high. Therefore, its use is only justified on large file servers, mail servers or web servers.

​

Fully Buffered Memories

One of the strengths of this type of memory is its almost zero margin of error: a reading error of 1,142,000 years is estimated. FB-DIMM modules use serial bidirectional tracks, which pass through each memory module, instead of having individual channels that send information to the modules, a concept quite similar to the principle of operation of PCI Express boards (also of serial technology) .

 

​

​