IT based services are intrinsic to every aspect of daily life and we take them for granted – so much so that when they don’t work it is a big issue for the service provider. Service providers have responded to these expectations by trying to improve availability – the measure of how reliable a service is – typically by investing heavily in redundant hardware and network links. Experience has shown however that the actual causes of unavailability are frequently nothing to do with hardware and networking, and more to do with human factors and software issues. This should raise alarm bells in every CIO’s mind – am I spending money wisely and will it actually deliver the availability that the business is expecting?

This paper starts by looking at business requirements and how to assess an appropriate level of availability for a service. Getting this right is important – the costs can triple for a relatively modest improvement in availability.

The paper then draws on Atos Origin research into the root causes of unavailability and the things that could have prevented it. It builds upon the unsurprising but often ignored fact that defective processes and human error cause far more problems than hardware failure, and proposes pragmatic solutions.

It is our view that availability can be engineered and that simple models can be used to predict the availability of a service and ensure that investment is targeted appropriately. The paper shows how modelling can be used to analyse a service and present alternative scenarios.

Throughout the paper it is acknowledged that different services have different availability requirements and that a proportionate approach is needed. This is depicted graphically showing the levels of availability at which each approach becomes viable.