Pipeline Publishing, Volume 4, Issue 3

	This Month's Issue:
	Automation

Autonomic Networks - Autonomic Communication

article page | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |

Decide: process with the best available computation engines including decision theory; risk analysis, hypothesis generation, genetic algorithms, and neural networks; scope actions to situations and needs.
Act: inside transactional contexts: invoke services; assemble components; manage collaboration; configure managed elements; signal devices; inform users and administrators; log actions and the strategic analysis that determined them.

Elements, Components & Services

There is some strong consistency of viewpoint on the major component elements of the AC. Today, many of these components are present in the standardized architecture of grids. The fundamental components of AN are:

Network/devices/sensors: the physical & electronic world
Modeler: a logical image of the network and communication paths linked via management control interfaces to the physical and configuration information of the network and maintaining a near-real-time display of state. Multiple models can exist with different purposes – for instance as load processors and predictive models of the future state. FineGrain Networks architecture has the modeler as a ‘living service’ always present in memory as an active computational model.
Registry: contains accessible information that stays relatively constant including all the physical maps, locations, and configuration baselines of all devices and services. Also contains user profiles and security credentials. Domain security contexts and service availability features are stored. (The 3GPP HSS is a step in this direction, as is the VeriSign security registry.)
Policy Engines: contain the rules and procedures for actions and activities.
Control systems: Invoke actions based on stimulus and maintain the overall health and stability of the autonomic network.

This architecture is radically different from today’s management structure. Device MIBs, agents, element managers, managers, manager of managers, service managers, etc. are no longer the principal management

machine. Instead, management is contained in the event-invoked actions of the control systems and is governed by state, policy, and situational context.

But we do not expect to throw out everything we have and have learned. Today, IMS is a policy-controlled architecture, and when it becomes incorporated with grids, will form bridging steps in the direction towards autonomic networks. What we expect is that the structure of management and the current instantiation of big, external management applications will become a marginal activity or disappear. When automation fails, there will still be a NOC with engineers who can respond to the unexpected or overly complex; who can resolve policy conflict. Contact Centers will still deal with issues and maintain CRM systems – for some time.

Our recommendation is that the current best thought and practice be encoded in agents and policies, becoming the brain of the initial central nervous system. For example, eTOM can be patterned into policy engines and control system logic.

But ‘small is better’ will win the day. Most decisions are very localized in scope, even today. A single device has a small configuration change. A bit of user profile data is updated. A message is sent. These kinds of actions can be realized with simple and local agents – they do not need the overhead of massive OSS applications. When small, when the decision is made in close proximity to the scope of its influence, massive parallelism becomes “built in” to the autonomic network. Scaling which would choke today’s applications is possible.

Similarly, as is the case with NGOSS, Framework services and Business Services will be present. Repeatable, consumable, commonly used services will be available as resources in the autonomic network. These framework services will be commodity items used at commodity prices. Business Services will be localized and individualized by company with more value-like pricing.

Finally tools for analysait, testing and validation will be framework services invoked as a normal part of service provisioning or restoration. When a device is installed and turned on it will discover the control network, register itself, seek and download its most recent software and configuration files. Then it will perform self-diagnostics, test all outside links and circuits, and flag itself as available. Similarly, software services and components will dynamically load into the business or control grid, instantiate themselves, register with control and discovery, download their configuration files, and automatically discover and link to all the resources they need. The specifics of this vary, the patterns alter, the

article page | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |

© 2006, All information contained herein is the sole property of Pipeline Publishing, LLC. Pipeline Publishing LLC reserves all rights and privileges regarding
the use of this information. Any unauthorized use, such as copying, modifying, or reprinting, will be prosecuted under the fullest extent under the governing law.