Hardware > Servers
Networks > Getting A Network

Server Hardware: What to Look For

By Morgan Killick

In the second of a three-part series starting with So You Think You Want a Server?, we look at the options for Server Hardware in more detail. An article on Server Software concludes the series.

Introduction

So, you have taken the decision to purchase a new or replacement server. How do you know what to look for? Who should you buy from? What type of backup system should you use?

The most important question here can only be answered by you, the buyer: how crucial is this server to my organisation? Servers can perform many different roles, but are all of these roles going to be performed by one piece of hardware? If so, or if you are running a fairly large network (15+PC’s), it will need to be of a higher specification than if you plan to split these roles. If the latter is the case, or for very small organisations (under 10PC’s), perhaps you can make some savings and stick with a lower specification system? Throughout this article we make reference to these ‘Higher’ and ‘Lower’ specification benchmarks as indicative examples.

Buying a server is not like buying a PC. Headline prices in magazines and websites show only a part-built system. The reason for this is that servers need to be customised according to particular needs. We look now in more detail at the main variables:

Variable 1: Chassis & Form Factor

The first and most basic choice is the server chassis and form factor. Easiest of all is the form factor (shape) of the server. Here you have a choice of ‘tower’ (standing upright) or ‘rack-mounted’. Rack-mounted servers can only be considered if your organisation has a 19” data cabinet. If not, a tower server can fit in wherever there is a suitable space away from risks.

By ‘chassis’ we mean the main parts of a server – the case, the motherboard, the types of connector on the motherboard and the expansion options.  Here the choice really boils down to the type of connector needed for your chosen disk technology (either SATA, SCSI or SAS – see below) as well as the size of the case in terms of future expansion. Broadly speaking, there is not much spare capacity in a rack-mounted server, so if you can see the need for future expansion, the bigger the tower, the better.

Variable 2: Disk Technologies

The primary function of most servers in small organisations is to store files and have them available to all network users. Since a server is likely to be accessing these files with much greater frequency than a PC, it makes sense that the server Hard Drives are as reliable and quick to access as possible. Indeed, the size, type and configuration of server Hard Drives are probably the most important factors in acquiring a server.

Serial ATA (SATA) is commonly used in ‘entry-level’ servers because it is cheap and has proved to be a great success in PC’s. However, the SATA disks seen in such servers are slower and less reliable than their cousin, SCSI (pronounced ‘scuzzy’). SCSI, at nearly 5 times the price, is expensive, but is considered essential for high performance systems. The latest version of SCSI is SAS - Serial Attached SCSI - essentially a SCSI drive but with a different connector allowing for improved performance in multi-disk servers.

The following table offers a quick comparison between SATA-150, SCSI U320 and SAS, the three main disk options provided by a typical server manufacturer in March 2007 

 
Certainly then, SCSI / SAS costs more per GB than SATA, but if your new server has a significant role in file storage and/or is the only server on a network of more than 15 or so PC’s, SCSI may be the better option. If you are considering a server with very heavy duties or with multiple hard drives, SAS will give the best performance but no innate improvement in reliability over plain SCSI

A further consideration is the number of Hard Drives to install in your server. For any server with a major role or roles, a RAID Array (Redundant Array of Independent Disks) should be considered vital. RAID means that the server has more than one disk in it to protect against disk failure. RAID is NOT a backup system! Rather then, that two or more drives mirror each other, so that if one develops a hardware fault, the server is able to carry on working.

Confusingly, RAID 1 is where two disks are used and RAID 5 is where three are used. RAID may either be controlled ‘onboard’ (i.e. as part of the server motherboard) or via a RAID card (an add-on, costing more money). The table below compares some typical RAID configurations:

Finally, with all these disk types, one has the option of ‘hot-swappable’ Hard Drives. These drives stick out of the front of the Chassis such that if they break, or the RAID array fails, a new one can be added without turning the server off. Hot-Swappable disks are needed if it is critical that a server is not turned off.

Variable 3: Processing Power (RAM and CPU)

The workload you have in mind for your server will govern the choice of RAM and Processor. If the server is just used for file storage, the workload is fairly low. If however the server is also acting as a mail server, intranet server, database server etc. it will need far more memory (RAM) and processing power.

RAM governs how many things a computer can do at once. RAM is relatively cheap to acquire and easy to integrate into your system. Ultimately this means it can be added at a later date if necessary. At the time of writing, 1024Mb (GB) should be considered the minimum for a Windows Server. As a rule of thumb, add another 512Mb for each major role the server is supposed to perform. It will only cost you an extra £60 or so but will speed up the server across the board.

Processor power governs the speed at which the server can perform complex tasks like finding a specific record in a large database. Modern processor performance usually outstrips the complexity of the tasks they are likely to perform. The type of processor in your server is therefore more likely to be decided by the manufacturers choice of motherboard (see Variable 1: Chassis) than by anything you can select. A typical lower spec server should have a Pentium 4 2.8Ghz or above. If you are asking your server to perform a lot of roles or are running a large office system you might prefer a Xeon processor, for example a Xeon 3.2 Ghz. For maximum upgradeability, a dual-capable motherboard (where an additional processor can be installed at a later date) is also an option to look out for

Variable 4: Backup & UPS

Contrary to popular belief, a backup system is not really a method for restoring an entire server to a previous state. Rather then, a backup acts as an archive of files (or email if running a mail server) such that if files are deleted or the server is stolen, there is a copy available to restore from. Backups ideally run overnight and are stored off-site (in the event of fire or theft).

There are three main types of backup system to choose from.

• Online backups are ‘software only’ and occur over the internet each night.
• External Hard Drive backups are becoming more popular as they offer high capacity, are quick to restore from, and are easy to get data from.
• Tape based backup systems (tape drives) however, are still the most common form of backup as they offer a highly portable and scalable method for retaining data from weeks if not months ago.

It is important to recognise that no backup can run without some software and regular attention whether from you or your IT Support provider. Tape systems in particular have a high maintenance overhead.

As far as deciding between the two hardware alternatives goes, the main determinant is your particular needs for archiving. If you need to store archives dating back weeks and months, the tape drive is for you. If on the other hand you just need some simple protection for the last few days’ worth of data, consider an external hard drive system (it will need a minimum of two drives to make it truly off-site). Cost-wise, external hard drives are around £90 each. Tape drives themselves are £200-£700 and cartridges are £5-£35 each.

Finally, an Uninterruptible Power Supply (UPS) should be considered essential. A sudden withdrawal or increase in the mains power to a server can have catastrophic results. A UPS then is a giant battery that regulates the flow of power to your server. Typically a UPS might keep a server turned on for 10-15 minutes during a power cut and if the power is not restored, gracefully shut it down. APC have a useful configurator that allows you to determine what size UPS (measured in VA) is appropriate for your needs. Expect to pay £200-£400 for a good UPS.

Manufacturers & Suppliers

So, you should have an idea by now of what kind of hardware you are looking for. But where to source from? Ideally you should go through your chosen installer as they should take responsibility for ordering the correct server configuration and should work with you to take into account your needs. Branded servers from the big four Manufacturers (IBM, Fujitsu, Dell, HP) are the mainstay of the server business, but there are a handful of smaller brands (e.g. Novatech, Pars, Supermicro), as well as a myriad of IT shops or Support companies who may well build a server from the components you specify.

Dell, in selling direct rather than through a network of resellers, often represent good value for money. However you may find that delivery charges, long delivery times and the ‘choose it yourself’ configuration routine are obstacles to getting what you want. HP, IBM and Fujitsu servers on the other hand are pricier but more readily available through knowledgeable resellers and system builders and this can mean a greater likelihood of quick turnaround as well as assistance with purchasing and someone to call if things go wrong with the order.

Warranty and Maintenance Considerations

One of the main advantages of going for a ‘big four’ branded server is that the manufacturer is likely to have a network of engineers ready for warranty repairs. It is important to recognise that these warranties cover components only. If your servers’ Hard Drive breaks, the manufacturer will replace the disk free of charge but will not reinstall Windows or recover your data!

Warranties usually come as either RTB (return-to-base); time-to-site; or time-to-fix. Time-to-site warranties will simply guarantee that an engineer will be at your site within this time. Treat with caution and do assess the cost properly! Although a manufacturers warranty will mean that you will not have to shell out for broken parts, you will need to identify an ICT Support company that can look after the server for routine maintenance as well as disaster recovery. This contractor may even be prepared to take on the responsibility for the hardware itself. Although you can minimise the risk of failure by following the advice in this article, there is a risk nevertheless. Servers are often ‘mission critical’ to an organisation once in place, so you will need to be sure you have support on hand should the worst happen.

Conclusion

The ‘How much does a server cost?’ issue is an open-ended question, involving many variables. The biggest costs in a server project are often the software, installation and support, which are not covered here. This article though should have guided you through some of the major decisions involved in acquiring the right hardware and given you an idea of the specification that you need.

Contained in the grid below is a comparison of our typical ‘Lower’ and ‘Higher’ Specification servers factoring in all of the variables discussed above.