Justin Sheehy says increased access to distributed systems has been the most interesting development over the last few years. He also suggests we should assume that any distributed system will eventually fail and we should have a plan to address that failure. The manner in which systems fail is changing. An advantage of increasing systems complexity is that individual component failure has less impact, but occurs more frequently. The result is systems that degrade slowly rather than quickly which provides a means to determine when a system is failing and allow more time to prepare for recovery.

Distributed systems have been changing. We now have service providers like Amazon and other cloud resources and the emergence of more general purpose distributed software. Meanwhile start-up costs are reduced while software and expertise have become more accessible.

In the next five years more people will be exposed to distributed systems and such ideas as cloud computing. One result will be increased pressure to connect systems and components that software developers haven't previously considered.  So our systems will become more complex, more accessible, less expensive, still subject to failure, but with an increased number warning points.

We are already beginning to see actual declarative approaches to systems which incorporate recovery protocols. An example is the Boom project in Berkeley by Joe Hellerstein and his students. They write declarative system specifications that can be run and subjected to formal analysis. This allows us to get the same kind of assurances about a distributed program that we've been able to do with local programs. Work in that general direction will have a significant impact by producing highly distributed, fault-tolerant environments in the cloud.