Rants
Questions
Soapbox
Best Practices
Apply today for a FREE subscription to CIO Magazine!
Sun, Oct 14, 2007 11:59 EDT

|
Posted by: Fred Hapgood in News Topic: ApplicationsBlog: Tomorrow's Buzz Today
Current Rating: |
Every CIO feels in their bones that networking management as we know it today is headed over a cliff. In the good old days, a decade or certainly two ago, networks were comprehensible; you could understand their parts and how they worked together, and, because you did, they were controllable in a very basic, mechanical way. You could turn an input up or down and have some idea of what would happen.
These days are passing. Utilization rates, numbers of nodes, and application types have been increasing every year, sometimes exponentially, always significantly, and the trend shows every sign of continuing until retirement. The center of networking is moving to mobile networks, often organized ad hoc, with continuously changing populations of players, and massively distributed WANs. Networks are becoming steadily more chaotic (in the technical sense that the size of a change and its consequence has no correspondence) and therefore harder to control.
Everyone's first thought about the right way to fix this problem is to retain today's general outlook and just automate it, allowing a far higher density of points of control. This approach, sometimes called autonomic or self-managing networking, has its appeal and is being intensively explored at many labs around the globe. However, it is unclear how successful automation can be given in the context of current philosophies of network operation. Conventional networks, automated or not, do not scale gracefully. This means that as networks get larger and more complex their management problems will keep getting hairier, the scale and costs of potential disruptions will grow, and the amount of time given to solve those problems shrink. Eventually these demands will crush any given implementation of automated management. It is, however, not clear what the alternative might be.
In this context some researchers have noticed that biology is rich with large networks -- protein cascades, gene switching networks, intercellular networks (of which the nervous system is only the best example), social networks among individuals, and ecological networks -- and seems to have no trouble extracting wonderfully adaptive and robust behaviors out of all these systems, even though none of these are managed in the usual (centralized) sense and are built out of parts with far more individual variation than our own networks could ever tolerate.
Recently researchers at the University of Bologna explored this thought by writing a network architecture modelled on a small number of biological processes. For instance, they developed a load balancing algorithm based on chemotaxis, a system used to inspire single-celled organisms cells to move in a given direction. If a cell needs to recruit other cells, it releases a hormone in pulses that radiate through the environment, like ripples in a pond. As soon as one wave of hormone reaches another cell of the right type, that cell starts to move in the direction defined by the point on its circumference that made first contact with the hormone and begins exuding more waves of the hormone itself.
Eventually, regardless of how noisy the environment is, and regardless of the ambient population level, the original cell will be able to recruit the cells it needs. The Bologna researchers used the same idea to do load-balancing: when parts of network find themselves stressed they emit signals that attract supplementary resources; the more load, the stronger the signal. Again, no matter what else might be wrong and regardless of scale, sooner or later the loads balance out throughout the network environment. The researchers have had comparable results using biological ideas to address other network
I think Fred Hapgood has this just right.
Austrian school economics and developmental biology have long swapped concepts and vocabulary to describe the development and behavior of as complex adaptive systems. I think we are, as Fred suggests, beginning to recognize networked engineered systems as complex adaptive systems.
Complex adaptive systems have large numbers of diverse agents that interact. Each agent reacts to the actions of the other agents and to changes in environment. Agents are autonomous, using distributed control and decentralized decision making, Eventually, the dominant interaction becomes the agents interacting with the system environment that was itself created by the agents’ own independent decision making.
In economics, we call the order that arises out of markets emergent self organization. In biology, we call it embryology. In either case, a large scale pattern emerges out of the smaller decisions and interactions. The emergent pattern is not imposed top-down, but rather arises from decentralized agents interacting within bounds of distributed control (or self control if you will).
A characteristic of meta-systems (or systems of systems) that demonstrate self organization is resilience in the face of change, what the economists call adaptive capacity. Market design theory, in the news this week with a new Nobel Prize, is in part concerned with ensuring adaptive capacity.
I was fascinated by the allusion to work in Bologna, defining system interactions in biological terms. I have been in conversations about the design of the power grid wherein we have touched on whether agent interactions should feel more like haptics, and the agent interactions more like tropisms. It sounds as if they have gone beyond conversation. Fred, can you do me the favor of posting a reference?
We are just beginning to apply the concepts of biology and markets to aggregates of engineered systems. In nature, systems that have too many direct interactions become brittle, and break badly. The Cleveland Outage of 2001 could be described as such a shattering, with the cracks extending into Canada and the East Coast. Less control and more heterogeneity in agents may be what we need to acquire resilience in our engineered systems.
________________________________________
It is the theory that decides what can be observed." - Albert Einstein
________________________________________
Toby Considine
Chair, OASIS oBIX TC
blog: www.NewDaedalus.com
>Fred, can you do me the favor of posting a reference?
http://fhapgood.fastmail.fm/MISC/bisonpaper.pdf
Ray Kurzwiel's book, "The Singularity is Near" is an excellent compilation of data on this and other related topics demonstrating the convergence of technology and biology. A facinating read, albeit long!
At its core, Autonomic Networks change our whole concept of ‘what is a network’, ‘where are the corporate boundaries’, and ‘what is network and service management’ - radically. Some of these principles are so compelling (such as discovery and self-configuration) that we are implementing them today even as our incomplete understanding of them improves year over year. But to truly understand and manage our existing and future networks, we need to see them as complex, organized, evolving systems and develop a ‘top down’ approach to controlling the transition of the existing network toward more autonomics. To make this practical concept a reality, we need (1) a well understood picture of how things actually work, (2) a set of clearly explicated goals, and (3) a set of plans and processes which provide for a practical transition from what we have to what we can see will be better.
In two articles published this summer in Pipeline Publications (www.pipelinepub.com) we explored where research and development is today in autonomic networks and autonomic communications. Then we dove into designing and building what we call self-* (pronounced self-star) systems. Self-* is a shortcut term for systems which are designed specifically to be self-organizing and self-managing, including properties such as: self-defining, self-configuring, self-awareness, self-optimizing, self-protecting, self-healing (self-monitoring, self-diagnostics, self-restoration). Further these systems exhibit the structural characteristic of self similarity. While we cannot yet exploit a full understanding of the behavior of autonomic networks we can see a way of exploiting the properties of autonomic communications. Self-* systems are ‘emerging’ today but it will take determined leadership to change.
You can follow this at Pipeline or download the white papers at:
http://www.ltcinternational.com/inside-out/2007/autonomic-networks-autonomic-communication.php
http://www.ltcinternational.com/inside-out/2007/self-networks-helping-networks-help-themselves.php