Mark Burgess (born 19 February 1966) is an independent researcher and writer, formerly professor at Oslo University College in Norway and creator of the CFEngine software and company, who is known for work in computer science in the field of policy-based configuration management.
Early life and education
Burgess was born in Maghull in the United Kingdom to English parents. He grew up in Bloxham, a small village in Oxfordshire from the age of 5–18, attending Bloxham Primary School, Warriner Secondary School and Banbury Upper School. He studied astrophysics at the (then) School of Physics at the University of Newcastle upon Tyne, where he later switched to pure Physics and then Theoretical Physics for his bachelor's degree. He stayed on to obtain a Doctor of Philosophy in Theoretical Physics (Quantum Field Theory) in Newcastle, in the field of Spontaneous Symmetry Breaking in Non-Abelian Gauge Theories, for which he received the Keith Runcorn Prize.
Burgess was invited to Norway for a two year Royal Society Post Doctoral fellowship in January 1991 by Professor Finn Ravndal of the University of Oslo, and stayed on for another two years funded by the Norwegian Research Council. While at the University of Oslo he developed an interest in the behaviour of computers as dynamic systems and began to apply ideas from physics to describe computer behaviour. He subsequently became the first professor with a title in Network and System administration at the same university. In 2023, in response to Brexit, Burgess applied for and became a citizen of Norway, following the acceptance of dual citizenship in Norway.
Burgess is perhaps best known as the author of the popular configuration management software package CFEngine, but has also made important contributions to the theory of the field of automation and policy based management, including the idea of operator convergence and promise theory.
Career
Burgess has made contributions to theoretical and empirical computer science, mainly in the area of the behaviour of computing infrastructure and services. In the early 1990s, Burgess asserted that programmatic models of computer programs could not describe observed behaviour at the macroscopic scale, and that statistical physics could be used instead, thus likening artificial systems to a quasi-natural phenomenon. With the increasing interest in the role of information in physics, Burgess has argued that computer science and physics can be bridged using the concepts of promise theory, through the notion of semantic spacetime, a description of functional aspects of spacetime at multiple scales, which offers an alternative to Robin Milner's theory of bigraphs.
Configuration
In 1993, Burgess introduced the software CFEngine based in intuitions and practice, focusing on the idea of repeatable desired end-state 'convergence', to manage system configuration. The term convergence, used by Burgess, is now often inaccurately just called idempotence, as convergence in his meaning implied both desired end-state and idempotence of an error correction operator at the desired end-state. Shifting interest from Theoretical Physics to Computer Science, Burgess then began to explore the ad hoc choices initially made, and set out to find a scientific method for understanding such choices in computing systems.
Computer immunology, anomaly detection, and machine learning
Following a position paper 'manifesto' pointing out the research challenges needed to make self-repairing systems, Burgess undertook to study computer systems as a number of empirical phenomena, taking an approach based on physics to learn first about the scales and patterns. The idea of self-healing, or self-maintaining systems was originally referred to as Computer Immunology, as it was inspired by research into the Danger model of human immune systems. The empirical studies were published in various formats between 1999 and 2003, culminating in a journal summary review, and a more practical method for automated machine learning of system behavioural characters. This incorporated the idea of so-called exponential smoothing (which was called a geometric average) for fast learning, along with a two-dimensional, cylindrical time model which was based on the result that network client-server traffic would be expected to behave like a quasi-periodic stochastic function (a characteristic of a system driven close to equilibrium).