Buying appropriate hardware is one of the most challenging parts of any IT manager's job. The benefits of keeping costs down, avoiding company-wide bottlenecks and properly maintaining hardware have to be carefully measured up against each other. Neglecting any one element is going to lead to problems, even if they're just with accounting.
For a server, scalability, efficiency and connectivity are the most important factors for avoiding false economies. When budgeting, ‘bang for your buck’ in terms of raw performance figures generally comes in second place on the priority list after these three areas.
Here is a brief comparison of some budget ranges weighed against the heavier hitters.
Entry-level servers: £300 - £1000
You can pick up a micro-tower server for as low as £300, while even the most minimal entry-level rack or blade server will tend to cost towards the high end of this range.
Rack or blade entry-level servers are intended to fill spaces in a specific existing setup as cheaply as possible, so rather than coming with built-in HDDs you are expected to use network-attached storage or external hard drives. They scale more easily and tend to be much more worth the cost as you acquire more and more.
Entry-level micro-tower servers are mostly designed around internal needs, with the assumption that they're going to be used for low load tasks in a small business environment. They're not ideal for use as web servers, especially if you expect or fear high loads, and using many at once is going to cause serious problems with efficiency and cooling.
HP offers a great variety of entry-level servers, where some other computing giants don't cater to an entry-level range.
Mid-range servers: £800-£3000
As with entry-level, a mid-range tower server will generally be cheaper and less suited to use at high scale than a mid-range rack or blade server. On the other hand, it will be much more cost-effective as a discrete unit. Bear in mind, though, that storing mission-critical data on a single tower server's single hard disk drive is inadvisable unless you also keep a back-up of this data somewhere else, or use a Redundant Array of Inexpensive Disks (RAID).
At this range, you start to get interesting choices to make; do you need more sockets or more processing power? More reliability in individual parts or safety in redundancy? Are you specifically going for energy-efficiency, or are you expecting to opt for more raw power to deal with random usage spikes?
Any server in this price range is going to handle most office jobs with ease, and a couple strung together with a load balancer should be able to deal with very high web traffic with a little bit of tweaking and care. You'll be able to handle various streaming tasks, including internal video-calls, at this sort of level.
If you need more than two or three servers, you should start to look more closely at rack or blade systems.
High-end servers: £2500-£5000
In this range of server you tend to get really strong all-rounders. These are the servers to get if you need excellent reliability, great processing performance and good scalability.
This is simply because the relative cost of adding, say, processing power starts to increase dramatically past a certain point – meaning that even if reliability is a distant last on your list of priorities, it's more worth it to invest a relatively modest amount to bring the reliability up to a decent level than it is worth to keep piling on the processing power.
The most expensive and powerful tower servers cluster around the bottom of this price range. Aside from very specific needs, this is the point at which most users will be shifting away from the tower form factor towards rack and blade servers, which are better-suited to use en masse.
If you want your servers to lack for nothing, and your budget is robust enough, this is the price range you'll be looking at.
Specialized servers: £5000-£50000+
The most sophisticated servers are powerful, specialized machines, usually aimed at a very specific market. For instance, high reliability, fault-tolerant servers are typically aimed at companies for whom more than ten minutes of down-time per year can be disastrous. On the other end of the spectrum, extremely high performance servers are often a necessity for complex engineering and scientific calculations.
It's rare that servers this specialized come in tower form, with most of them being the very highest-end of blade servers. This is because it's rare that specialized servers will be used alone. It's simply less safe and far too limiting in terms of power. If you're optimizing your server build purely for those two aspects, you're going to need more than one machine.
Network hardware tends to be relatively straightforward to order. You order the best, most reliable, most long-lasting equipment that meets your requirements.
When buying switches, one of the most important limiting factors in terms of performance is the amount of memory. Cheaper switches may support important features such as Store and Forward mode, but they will usually have small amounts of memory buffer for each port – often between 256 and 512 kilobytes. This means that you don't have enough memory to prevent packets being discarded, and you'll take a huge performance hit as soon as you step beyond two computers communicating using the switch.
Switches should also support spanning tree protocol to enable loop avoidance. Fortunately, all modern routers and switches support STP, but for refurbished older switches you should do a little research if they are more than a few years old.
Again, there are many ranges of hubs and switches available. Hubs are rare since switches have fallen in price to the point where the advantages of a hub are very rarely worthwhile.
Routers are primarily divided between Small Office/Home Office (SOHO) and Enterprise classes (although we recommend against searching for "Enterprise Bridge" unless you're particularly interested in Star Trek).
For SOHO use, key terms to look out for are:
Modem: A DSL modem built into your router provides the basics of internet connectivity, but not usually much more.
It is capable of being remotely managed using telnet, although this is not recommended due to the potential for man-in-the-middle attacks.
The main difference between a modem and CSU/DSU (which explains why a modem can be managed with telnet) is that a modem converts digital signals so they can travel on a phone line, while a CSU/DSU converts digital signals to travel on a leased line.
CSU/DSU: Channel Service Unit and Data Service Units are packaged as small pieces of hardware (about the same size as a modem – no coincidence, as at the highest level of abstraction their functions are similar) which allow your local area network to communicate with a wider network – such as the one your ISP offers.
If you lease a line (which you may know as a T1, E1, T3 or E3 line), you'll need a CSU/DSU.
A CSU will provide extra features such as Line Protection (protecting your network from power surges, such as lightning strikes) and some amount of in/out equalization of the signal. The DSU will perform the signal conversion.
They should also provide some remote testing capabilities, which can be very useful.
CSUs and DSUs can come separately, but it is most likely that when buying a SOHO router it will come with a combined CSU/DSU unit.
Quality of Service Features: These features are for optimizing voice and video applications, and may be invaluable if you rely on voice or video streaming for any aspect of your communications.
Gigabit Ethernet: While almost all routers will support Ethernet and Fast Ethernet, which offer speeds of 10 Megabits/second and 100 Megabits/second respectively, not all SOHO routers offer Gigabit Ethernet support. As the name suggests, this form of Ethernet provides speeds up to 1000 Megabits/second, an order of magnitude improvement over Fast Ethernet.
For a very small office, it’s unlikely that you will need Gigabit Ethernet unless you are transferring very large files on a regular basis from computer to computer. For example, if you’re involved in graphic design, animation or video work you should look for Gigabit Ethernet support.
For enterprise use, routers will generally need to provide superior performance in the form of ten Gigabit Ethernet. Whilst other typically necessary features will include flexibility, support of multiple technologies, as well as very high levels of redundancy and resiliency. For those looking for a one-stop solution, a BIG-IP is the only device within the industry that can do all of the above, whilst being adaptable to current business and technology trends too.
Desktop hardware for business tends to be standard equipment. You'll really not be expecting to use large amounts of processing power at work unless you're working with demanding applications such as 3D design tools.
There are significant issues to bear in mind though.
First and foremost, you need to commit to a use case. If you're going to need to use multiple resource-intensive programs at once, commit to that vision completely and design the other specs you need around that use case. If this means you can't fit within budget, it's better to buy fewer PCs that will be capable of meeting your demands than more PCs which won't. This is easy to say, but hard to adhere to in practice, as it can make managing who gets to use which PCs when difficult.
Secondly, almost as an inversion of the use case, there's the question of limits. If you're sure you're never going to use more than 2GB of RAM, there's no point spending money on more "just in case". Stick to your limits, as it could mean that you can afford a machine that's more powerful in other ways (say, hard disk space or CPU).
You'll almost never need a powerful GPU (graphics card) for an ordinary workplace desktop. If you do need a GPU, you're better off opting for a high-end workstation, which will fit your use case much better.
Overinvestment as well as underinvestment will cost your business real money, and it's easier to get carried away with desktop hardware and order over-specced machines than it is with server hardware. Be careful with your money, and spend it where it counts.
Tablets & Peripherals
Tablets, notebooks and laptops are increasingly important to many workers who feel more comfortable working on them, and who may value the freedom to work in different locations or on the move.
‘Bring your own device’ policies can lead to serious security problems, especially as people find ever-more-inventive ways of bridging the 'air gap' between machines.
As a consequence of this (and more plausible security issues, like workers connecting their devices directly to your network) it's becoming necessary to use company-owned and company-controlled tablet devices.
Most tablets are much the same, so your decision will mostly hinge on whether you need a physical keyboard attached, and how much faith you have in the manufacturer of the hardware and software.
For hardware peripherals, you should look for devices designed specifically for business use. Consumer peripherals are not generally designed to integrate well with complex networks, and may be less reliable or be designed so that they will wear out after a relatively small number of uses.
Refurbished hardware is hardware which has been owned, but has been tested, cosmetically restored and reset to OEM factory settings. With refurbished network equipment failure rates are greatly reduced because it has been used and tested. Purchasers should check that any equipment they are looking at comes with a 12 month warranty, ensuring total piece of mind.
Refurbished hardware can attract discounts as high as ninety five percent, and is often a very good choice for cheap clusters of servers or bargain laptops.
Hopefully that's answered a few questions about getting the right hardware, if you didn't know before; or clarified the reasoning behind making the right hardware choices.
If you've got any further questions, please ask in the comments below!