CPM Report 08-198: System Farming - Bruce Edmonds

We discuss the implications of emergence and complexity for the management of complex distributed systems (CDS). We argue that while formal design methods may play a role, they have distinct limitations where it comes to complex systems. There are similar limitations to statistical methods. Thus we must look to other ways of managing these systems, involving a shift from: prior one-off design towards post hoc continual management; from predictive abstract theory towards detailed descriptive modelling to guide monitoring and aid diagnosis; from system optimisation to simple disaster prevention; from single models to many models; from single well-designed mechanisms to multiple overlapping mechanisms; from individual to collective effort. We call upon those in the SASO community to explicitly reject those tenets that are only useful with simple systems. In other words, when trying to understand CDS, become more like zoologists rather than mathematicians and when managing them becoming more like farmers than engineers (at least in the classic sense).

Available at: http://cfpm.org/cpmrep198.html