Management for IPbased Applications Mike Fisher BTexa CT

Management for IP-based Applications Mike Fisher BTexa. CT Research mike. fisher@bt. com

Introduction • • Future networks and applications Active Networks Management Problems Active Management

Evolution of the Internet • Demand for new applications/ customisation • Infrastructure can’t keep pace with application explosion • Multimedia – need for Qo. S, correlated flows, multipoint • Control of end-to-end application performance • Flexible infrastructure and adaptable management

Multi-owner Network Administrator 3 Administrator 2 Public Network Policies (SLAs) Administrator 1 User ?

Programmable Networks • computation in the network, not just routing • users can introduce programs – delegate control and responsibility – improved resilience to change – … increased risks from sharing control

Active Network Technology • Dynamically update software on network element to change node behaviour • Users/Operators/Value-Added Service Providers create new services to run on active nodes • Active nodes include routers, proxies, firewalls etc • Two categories: – capsule-based approach - packets may contain both data and active code to be executed at node – discrete approach - active code downloaded out-of-band from code libraries/caches

Locating Active Programmability Core network devices Host Border devices Edge devices

P 1520 reference model

Architectural Requirements • Divide programmability • Active Router – OSI layer-3 functions – embedded scripts or programs, from trusted sources – low memory and computational power • Active Server – application layer active networking – many specialised nodes • transcoding node requires efficient maths operations • node supporting active caching require high-performance I/O

Active Architecture

Active Virtual Network virtual space normal link active server router active router virtual network link client

Management problems • • • High percentage of IP VPN costs New features (e. g. multicast, Qo. S) do not get added Changing MIBs is extremely difficult Centralised control model limits scaling Inefficient information flows Manual intervention • Made worse by demands of new applications

A solution? • • • Minimise operator intervention Enable flexible addition of features Support diverse information models Use high-level policy-based interfaces Distribute and delegate Give responsibility to customers • Active Management System

Active Management • No central point of control • Autonomous decision making based on policies and local knowledge • Dynamically introduce – – new active server types new policy sets new mechanisms for policy storage and retrieval new algorithms for policy decisions and enforcement

Active Management • Hierarchical • Autonomous Management agent Admins Autonomous controller Users EEP Proxylets

Summary • Flexibility in services demands an active approach • Two levels of programmability – active server – active router • Approach to management – programmable infrastructure – delegate application management • Policy-based active management required