Policy interoperability and network autonomics

Abstract

Autonomic behaviours in network operations will alleviate much of the labour intensive and error prone interventions of today's complex networks. The Service Provider must be able to manage the infrastructure and services at an abstract level, focusing on what the desired behaviour should be rather than how it might be specifically achieved. Policy-Based Network Management (PBNM) appears as one of the leading mechanisms to describe desired behaviours and abstract the programmability of an autonomic network infrastructure to the Service Provider. For massive-scale and complex networks, the current understanding of the Higher Level to Lower Level (HL→LL) refinement process commonly used in PBNM today is not completely effective. One problem encountered is the need to provide a bind mechanism between Higher Level and Lower Level policy specifications such that cross-layer policy requests in the policy continuum can be made by lower policy layers in a dynamic policy refinement cycle (LL→HL→LL). In this paper, we illustrate the problem with a policy-based simple admission control (SAC) application. We then show that policy specifications with a join operator (times sign closed) simplify the SAC specification. We also investigate the performance considerations of this enhancement in Internet size applications. Our future goal is to provide a policy inference engine that can support complex specifications appropriate for PBNM systems that support autonomic behaviours in large networks, made of Network elements with realistic memory and processing constraints. © IFIP International Federation for Information Processing 2005.