The Problem

Web servers represent a number of challenges in their design and construction. The first and over-riding challenge is to ensure that they are reliable. Clearly, this sets challenges for both the hardware and the software. As the customer nearly always specifies the operating system this area becomes their responsibility but that still leaves the hardware issues.

Design Criteria

Within the over-riding constraint of reliability come the issues of compactness (often a rack mounted machine), capacity (the more disk and cpu capacity the better), heat dissipation (essential to continued operation and to the life time of the machine) and support. We at Tree of Life have developed many servers, both tower cased and rack mounted, to exacting standards. To maximise the life expectancy of the machines, great care has been expended in designing and implementing the heat extraction from the machines, even in crowded machine racks. Once again, the quality of the heat dissipation methodology dictates the number and speed of the disk drives the case will take in real world operation and also the CPU processor that may be employed.

Server Replication

When specifying servers there is a real and careful judgement to be made between the cost implications of, redundancy in disks (RAID arrays) and power supplies (hot swappable redundant power supplies) in a server, and entirely duplicated or mirrored machines. Overall replication of machines offers a higher level of security than redundant power supplies and mirror arrays of disks can ever achieve. Complete system replication has now become very economic because operating system cost can be controlled and very low cost powerful basic computer hardware is available. The cost of two machines, in hardware terms, is much more than offset by the cost of using a good quality RAID card and hot swappable power supplies on a single machine. If an expensive operating system is to be replicated, however, then the decision may well swing the other way.

Fully Resilient File Serving

We have had much success in developing replicated file servers to provide a very high level of resilience and security at a low cost. When that replication can be affected somewhere off-site then this removes the need for an expensive tape backup system and other off-site backups. By linking machines in widely separated locations and also locally we have achieved fully automatic, remote backup and hot running fail-over at a very modest cost. Using the Linux operating system has meant that the operating system costs, in fact all software package costs, are negligible and what is involved is suitable network connections and the effort required to set up the arrangement.