Does it strike you odd that, despite the outage of Amazon Elastic Compute Cloud (EC2) and Elastic Block Storage on April 21 2011, many of the customers who were affected are still so keen on the service?

It puzzled me until I considered the huge diversity of applications out there. If these customers had been trading eCommerce sites, 2-4 days of downtime would have been a severe blow. For a provider of financial services, even a day’s loss of services is unthinkable. The problems of the Millennium Exchange system on the London Stock exchange in February, for example, lasted only four hours, but caused a 30% drop in FTSE 100 turnover.

For other types of application, however, this isn’t such a problem, and herein lies the source of a major sticking point in the on-going debate about the usefulness of the ‘cloud’: we all extend our limited experience of commercial IT and generalize it beyond reason. We think that what was good for the projects we were directly involved with would be good generally, but it’s simply not true.

As the IT industry diversifies from its roots in commerce, taking on social interaction, gaming, gambling, armaments, manufacturing, espionage, publishing, politics, and so on, it becomes ever more fragmented.

A generation of developers, the ‘Dotcom-boom-babies’, has been able to work in IT with far less experience of the constituency where current dogmas and ‘best-practices’ first evolved. What is appropriate for developing applications that are critical to the health of international corporations, under the yoke of international legislation for handling and storing financial and personal data, would be curiously arcane for a ‘fluffy’, rapidly-evolving networking site. An architectural framework that suits a site like this is worlds away from the resilience required for a system that supports a major company;

Even within the corporates, there are huge differences that will determine IT architectural choices. The culture of IT in manufacturing is quite different from that in banking or the way that government IT is run. A manager of a major medical application once remarked to me that whereas a company might lose millions in revenue if one of my applications suffered more than a few minutes of downtime, he was kept concentrated on the fact that people would die if one of his applications failed.

The marketing men may bombard us with generalizations about visualization, the cloud, distributed architectures, agile development techniques, as if they were back to selling soap powder, but in such a diverse industry, one man’s meat is another man’s poison.