The dichotomy of Network and Services still exists, yet AC provides for collaborative automation of services in conjunction with networks. The goals and activities of each layer still are distinct; it is the patterns which repeat across layers. At some extremes of design, visionaries see networks as physically self assembling. These distant future networks would look strange to us. But most effort today is directed at logical self assembly and management. What is radically different is a profound situational awareness in each layer of: (a) the activities and (b) the structure of other layers and (c) the environment. In fact, the amount of information available when testing the truth of a firing rule or in determining behavior in near-real-time, may be prohibitively large – negatively impacting performance. Context filtering will go hand in hand with context awareness.

Nevertheless, we expect many past lessons of network design and management must be encoded in the new architectures. For instance, not everything needs to be dynamic all the time. When covered by automatic restoration, lower levels can be “nailed up” for longish time periods, letting the higher levels adjust and adapt in shorter periods. Also, it is not always necessary to provision a new VPN if a data flow can identify and then hitch a ride on an existing virtual circuit with the necessary QoS and Security. But in AC, no longer would the layers compete at network restoration: a collaborative agent would invoke restoration according to the fit of pattern at that specific event circumstance – for example, speed-of-reaction still overriding for extreme outages.

Fundamental to the design of all autonomic networks is an embedded security model. In traditional networks, security was applied after the fact - mostly using external devices, applications and tools. The world would be a much more organized and safer place if TCP/IP had been designed with embedded security – and even this is being looked at again. Autonomic networks must be self configuring, self deploying, and self assembling. Without iron-clad AAA (Authentication, Authorization, & Accounting) built into this structure, these autonomic service building blocks are just virus. While absolute security cannot be guaranteed and security does add cost, most current designs have produced workable compromises where the security of a network is well known. The best designs can adjust cost and security as properties of network domains, controlled by policy and bounded on the borders with a AAA that filters user and service admittance.

Several more characteristics of AC must become common features of devices and services. These are novel behaviors not generally present in today’s networks; except for a few in some specific instances such as

Cisco’s proposed Application Aware VPN (AAVPN) and Ipanema’s commercially available edge controller. Situational awareness covers the capability of services to autonomously adapt to the context from which they are requested and in which they execute. It demands that technologies capture situational data and effectively exploit it. Self-organization uses control services that implement patterns often derived from nature and identified in the extensive examples and principles of Complexity Science. Fractal/self similarity concentrates on patterns of organization that reproduce nearly identical structures over multiple scales. Very similar to the concept of recursion found in the definition of for example the Pascal programming language, we see this embodied in the fractal graphs of complex systems. In AU, networks should organize using similar patterns at every layer and in every sub-domain. Only the mix of components and services and the goals change from sub-system to subsystem and layer to layer, not the structure of organization. Self-healing/ Self-preservation/ survivability which provides a way for every device, path, and service to recreate itself upon failures, ameliorating the impact of the failure on overall service delivery. [more in the next issue] Traditional network architectures have been largely concerned with structure – what is where, what connects to what, what protocols are present. Architecture in AC is more concerned with behavior and process than with structure. Since the network is expected to have situational and context awareness, the architecture looks like a central nervous system performing its reactions to stimuli. Four basic stages in this reaction (per Dobson’s ‘A Survey of Autonomic Communications’) are designed into AN: Collect: gather information from environmental and device sensors and instrumentations; determine application or service requirements; look up user context and security credentials. Analyze: use the most modern of inference tools to process this collected information including: Fuzzy logic engines, bounds and envelopes, economic models, rules and polices, and game theory. article page | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |