Networking Research

Autonomic Networking

By: Brian Carpenter

Date: November 1, 2014

line break image

IETF 90 included a well-attended Birds of a Feather (BoF) meeting entitled, “Use Cases for Autonomic Networking (UCAN).” The associated mailing list is called ANIMA, which stands for Autonomic Networking Integrated Model and Approach. So what exactly is autonomic networking?

The dictionary recursively defines autonomic as “relating to, affecting, or controlled by the autonomic nervous system,” which doesn’t help much. The autonomic nervous system is an important aspect of an animal’s body—it takes care of vital functions, such as breathing and swallowing, without conscious control. This is what inspired the concept of autonomic computing, which was introduced by IBM in 2001 with the intention of making computing systems as self-managing as possible. Wikipedia explains it in more detail: “Autonomic computing refers to the self-managing characteristics of distributed computing resources, adapting to unpredictable changes while hiding intrinsic complexity from operators and users.” Autonomic networking, which has been an active area of research in recent years, refers to the application of such ideas to networks. One forum for these discussions has been the Network Management Research Group of the Internet Research Task Force (IRTF).

Autonomic Networking (AN)

One way to look at autonomic networking is “plug and play for the ISP” or “plug and play for the enterprise network.” This is a step forward from the original concept of plug and play for home networks, which has long been recognised as a vital requirement (see, for example, the work of the IETF HOMENET working group).

The goal of self-management includes self-configuration, self-optimization, self-healing, and self-protection. AN puts operational intelligence into algorithms at the node level in order to minimize dependency on human administrators and central management. Nodes that participate in AN discover information about the surrounding network and negotiate parameter settings with their neighbours and other nodes. Ideally, autonomic nodes use stable closed-loop control methods to achieve self-management, instead of using more traditional top-down network configuration and monitoring tools to set and verify their parameters. Nodes may also have learning and cognitive capability, including the ability to self-adapt decision-making processes based on information and knowledge sensed from their environment. In the most sophisticated cases, advanced data analytics may be part of the input to the autonomic mechanisms.

More than Science Fiction

Many aspects of small networks have been self-configuring for years, including unmanaged home and small office networks. And numerous existing protocols have a flavour of autonomic properties (e.g., the spanning-tree algorithm needs no manual configuration in order to operate, and some routing protocols require very little configuration). Recently, prototypes and initial products of explicitly autonomic protocols have emerged from some of the major networking equipment vendors. However, it is clearly necessary to have some basic standards in place if AN is to become relevant to large multivendor networks.

Why Now

The main motivation is not new: large network operators, both Internet service providers (ISPs) and enterprises, have been increasingly suffering from the problems and difficulties caused by the central configuration of hundreds or thousands of network elements. Now, after years of research and discussion, ideas about how to achieve autonomic networking are becoming concrete. Fortunately, it is now also economic to provide enough computing power and memory in network elements to support AN. The time is therefore ripe for a standardisation effort.

Network Parameters

A number of use cases for large networks were proposed at the UCAN BoF: network address and prefix management, optimisation of mobile backhaul links, risk-aware routing, and detection of service-level agreement (SLA) violations. Other examples are starting to emerge as well, including monitoring and reporting, and others will certainly follow. Two very fundamental aspects of AN can be viewed as use cases in themselves: securely bootstrapping new devices, and creating a secure autonomic control plane for use by specific AN applications.

Keeping Control

While it is obviously desirable to reduce the need for tedious human interventions, it is essential that network managers can ensure that the network does what is needed and remains fully secure, even if many nodes are configuring and managing themselves. For this reason, the model for AN must include a mechanism for communicating the intent of human managers to all self-managing nodes for matters such as resource control, service requirements, and security policy. At the same time, in real networks, AN mechanisms will need to coexist with traditional top-down management and monitoring tools for many years, so it must be possible to introduce AN technology in small steps.

Next Steps

At the time of this writing, an ANIMA working group is under discussion in the IETF. A complete solution for autonomic networking would be a very ambitious goal. The scope of the proposed effort is much more modest: define a minimum set of specific reusable infrastructure components to support autonomic interactions between devices, and specify the application of these components to one or two elementary use cases of general value. The main goal is therefore to develop common infrastructure components for distributed functions. The infrastructure should be capable of providing the following services to those distributed functions:

  • A common way to identify nodes
  • A common security model
  • A discovery mechanism
  • A negotiation mechanism to enable closed-loop interactions
  • A secure and logically separated communications channel
  • A consistent autonomic management model

Some important topics are intentionally not included in these initial goals as they are considered separate matters that should be considered later:

  • Mechanism for distributing policy intent to autonomic nodes
  • Use of data analytics by autonomic nodes
  • Other external information sources
  • Systemwide integration of autonomics

Further Reading

draft-irtf-nmrg-autonomic-network-definitions and draft-irtf-nmrg-an-gap-analysis

Mailing list: anima@ietf.org

Acknowledgements: Several IRTF and IETF drafts were raided for text and ideas, and useful comments on this article were made by Leo Dorrendorf, Sheng Jiang, and Alexandre Petrescu.

No Comments to Show

Leave a Reply

Your email address will not be published. Required fields are marked *