Tower servers
Tower servers seem dated and look more like desktops than servers, but these servers can pack a punch. In general, if you have a lot of servers, you're probably not using a bunch of tower servers, because they can take up a lot of space and are tough to physically manage since you can't easily stack them on one another. In some cases as organizations grow and move to rack servers, conversion kits can be purchased to turn a tower server into a rack-mount server.As implied, tower servers are probably found more often in smaller environments than anywhere else, although you might find them in point solutions in larger places.
Tower servers are generally on the lower end price-wise, although they can expand pretty decently and become really expensive.
Tower servers take up a lot of space and require individual monitors, keyboards, and mice or a keyboard, video, mouse (KVM) switch that allows them to be managed with a single set of equipment. In addition, cabling can be no fun, especially if you have a lot of network adapters and other I/O needs. You'll have cables everywhere.
I don't buy a lot of tower servers these days, but they still have a place. My most recent tower server purchase was to serve as my backup system running Microsoft Data Protection Manager 2010.
Rack servers
If you run a data center of any reasonable size, you've probably used a lot of industry standard 19″ wide rack servers. Sized in Us (which is a single 1.75″ rack unit), rack servers can range from 1U "pizza boxes" to 5U, 8U, and more. In general, the bigger the server, the more expansion opportunities are available.Rack servers are extremely common and make their home inside these racks along with other critical data center equipment such as backup batteries, switches, and storage arrays. Rack servers make it easy to keep things neat and orderly since most racks include cable management of some kind. However, rack servers don't really simplify the cabling morass since you still need a lot of cabling to make everything work — it's just neater. I once worked in a data center in which I had to deploy 42 2U Dell servers into three racks. Each server had to have dual power cables, keyboard, video, and mouse cables and six (yes, six) network cables (six colors with each color denoting a specific network). It was a tough task to keep the cabling under control, to put it mildly. Because everything was racked, there was built-in cable management that made this easier.
Like tower servers, rack servers often need KVM capability in order to be managed, although some organizations simply push a monitor cart around and connect to video and USB ports on the front of the server so that they don't need to worry about KVM.
Rack servers are very expandable; some include 12 or more disks right in the chassis and support for four or more processors, each with multiple cores. In addition, many rack servers support large amounts of RAM, so these devices can be computing powerhouses.
Blade servers
There was a day when buying individual blade servers meant trading expansion possibilities for compactness. Although this is still true to some extent, today's blade servers pack quite a wallop. I recently purchased a half-height Dell M610 blade server with 96 GB of RAM and two six-core processors.There is still some truth to the fact that blade servers have expansion challenges when compared to the tower and rack-based options. For example, most tower servers have pretty significant expansion options when it comes to PCI/PCI Express slots and more disk drives. Many blade servers are limited to two to four internal hard drives, although organizations that use blade servers are likely to have shared storage of some kind backing the blade system.
Further, when it comes to I/O expansion options, blade servers are a bit limited by their lack of expansion slots. Some blade servers boast PCI or PCI Express expansion slots, but for most blade servers, expansion is achieved through the use of specially designed expansion cards. In my case, the Dell M600 and M610 blades have three mezzanines. The first mezzanine consists of dual Gigabit Ethernet adapters. The remaining mezzanines are populated based on organizational need. In my case, our blades have a second set of Gigabit Ethernet adapters housed in the second mezzanine and Fibre Channel adapters in the third. If necessary, I could also choose to use mezzanine cards with four ports in some configurations. So, although the blade server doesn't have quite the I/O selection of other server form factors, it's no slouch, either.
When raw computing power and server density is the key drive, blade servers meet the need. For example, in my environment, I have a 10U Dell M1000e blade chassis that can support up to 16 servers. So, each server uses the equivalent of 0.625U of rack space. On top of that, the blade chassis holds four gigabit Ethernet switches and two Fibre Channel switches, so there is additional rack space savings since I don't need to rack mount these devices to support different connectivity options. In addition, the blade chassis has a built-in KVM switch so I don't need to buy a third party and cable it up.
Speaking of cabling, a blade environment generally has much less of it than tower or rack environments since a lot of the connectivity is handled internally. You'll end up with a neater server room as a result.
Another point is adding a new server consists of simply sliding it into an available slot in the chassis. There is no need to rack a new server and deal with a bunch of new cabling. This small size makes heat dissipation a challenge. Blade chassis can put out a lot of heat.
From a cost perspective, blade servers require some initial infrastructure, such as the chassis, so the upfront cost is often higher than for servers of other types.
No comments:
Post a Comment