PC server vendors, such as Compaq and Dell, are currently working with Microsoft to develop a server clustering technology called Wolfpack, which should be launched during the first half of 1997.
The technology consists of linking together two or more servers to make the network more reliable. Clustering also improves scalability: if you need more power to support extra users or a new application, you simply add more servers to the cluster.
The basic concept underlying server clustering is redundancy. The idea is that a system has more components than it actually needs, so that if one component fails, a spare or redundant component can take over.
It is not simply a matter of keeping a spare Pentium chip or power supply in a cupboard - the redundancy has to be an integral part of the system. A useful example is a RAID (Redundant Array of Inexpensive Disks) system, in which a group of hard disks are configured so no data is lost if the system malfunctions.
It is increasingly common for state-of-the-art PC servers, such as the Compaq Proliant 5000, to have different forms of redundancy. In addition to RAID, there are redundant power modules. This means that if one processor fails, another can take over. Sometimes, even redundant network interface cards are provided in case the main network card fails.
In addition, software such as Compaq's Insight Management can check the system for potential component failures. This means that you can take appropriate action before problems arise.
But even with the best hardware and software, it is possible that the whole system could fail. This has led server vendors to consider server clustering.
But the technology is not simply a case of adding more individual servers to a network to deal with new tasks. With clustering, users see only one device and are not aware of whether it is a single server or a cluster.
Server clustering is a fairly complex operation because functions which normally take place in one server system unit have to be distributed over several units. So there has to be a close match between the server hardware and the network operating system (NOS).
At its most basic level, server clustering technology is nothing new.
Many companies have been running Novell's System Fault Tolerance III (SFT III) for years, first on Netware 3.12 and now on Netware 4.x.
The first thing you need is two identical servers which have been certified by Novell for use with SFT III. They must have equal amounts of RAM and storage, each with identical network adaptors, identical monitors and monitor adaptors, and a high-speed mirrored server link (MSL) for each server.
One server is configured as the primary server and the other as the secondary server. Both run Novell SFT III, and all the time the primary server sends a signal along the MSL, a fibre optic cable, which says 'I am alive and working properly'. It is called the heartbeat signal.
The servers don't need to be - and ideally shouldn't be - in the same location. In fact, they can be up to 40km apart, depending on the configuration of the MSL.
When the primary server's heartbeat stops, the secondary server automatically takes over. In many cases, this happens so fast on the network that users don't notice the interruption. The secondary server becomes the primary server and vice versa. Once the problem with the faulty server has been fixed, it then becomes the secondary server to the primary.
The latest version of Netware 4.1 SFT III can use the multiple central processors which are increasingly found in many servers. One processor undertakes all the hardware-related input and output tasks and the mirroring synchronisation. The other processor carries out file services and other server mirroring software services. As an extra precaution, a redundant MSL link can also be installed.
Novell SFT III is the simplest and most easily understood server clustering technology. Another server clustering technique is shared mass storage, so that when the secondary server takes over from the primary it has access to all the user data.
But the most efficient and complicated form of server clustering goes further towards full redundancy. It involves multiple servers working in synchronisation with one another. They all undertake the same tasks and store data on their own storage systems. If one server goes down, the cluster just loses a proportion of its total processing power.
Clustering will become much more common when Microsoft launches its Wolfpack technology over the next few months. First announced in October 1995, Wolfpack is based, unsurprisingly, on Windows NT Server.
The main part of Wolfpack is the clustering application programming interface (API). This allows software developers to write applications for Windows NT Server which can use its clustering abilities. Wolfpack will be delivered in two parts: the first phase provides a 'fail-over' ability so that if one server fails, another continues with its task , in the same way as Novell SFT III. With the second phase, there is the facility to connect more than two servers for improved performance and redundancy.
In many ways, clustering is an exercise in migrating mainframe functions, in terms of redundancy and scalability of power, onto a PC platform. The difference is that PC server clustering should be much cheaper to implement than a mainframe system.
A PC server cluster system is also what is known as a granular system: you can remove or add small parts as and when they are needed. It was not always so easy to upgrade components of the monolithic mainframe.
Also, rather than being locked into parts for a proprietary system, PC components are available from a variety of manufacturers, which ensures the best deal.
Centralising shared network resources on a server certainly makes sense.
But the data and services that are provided by the server are always at risk. Basic security measures such as controlling access, regular back-ups, an uninterruptible power supply (UPS) and constantly monitoring the server's components are fairly simple and cheap to implement.
However, more companies are realising that, with increased reliance on PC servers, they have to implement more complex measures which provide a networked system with a certain amount of redundancy.
Neil Martin of Panda Security discusses Epic Games' decision to avoid the Google Play Store in its Android release of its popular game Fortnite
Musk went public on privatisation plan "because I felt it was the right and fair thing to do so"
Intel's 9th generation Core CPUs will be released on 1 October along with Z390 motherboards
Short-sellers burnt by Musk's "false and misleading" tweets the first to file suit