Reliability of Complex Services

In the relatively young history of computers, their capability has grown at an exponential rate – both in hardware (see Moore’s Law), and in software. Naturally, along with the increased potential came the greater magnitude of complexity. Where mainframes used to run isolated instructions from single programs, today’s conditions require multiple interactions at every level – each CPU multitasks dozens of processes, servers manage their load between multiple CPUs, application tiers take advantage of server farms to manage load, service providers require multiple tiers for efficient services. An industrial-strength service architecture pushes the limits of descriptive reliability models, as the number of nodes and connections in dependency graphs of the service components grows rapidly. Without a doubt, the inherent complexity of service providers, whether it is a brokerage firm, an online retailer, a search engine or a portal, is significant. But on the other hand, the availability of the Internet and global access require these providers to be available 24 hours a day, seven days a week. Reliability being a major concern, much of the marketing literature for service components is filled with terms like “five nines”, signifying 99.999% availability – but in real life, we typically find that reliability is much less – anyone with a computer has grown accustomed to encountering various faults, in a manner not applied to older technologies, like land-line phone service or automobiles.