Architecture of co-designed network management in 4WARD

© 4WARD Consortium Confidential

Giorgio NunziNEC Laboratories Europe

Motivations for In-Network Management

� Limitations of traditional approaches to Network Management

– Network infrastructure is deployed first

– Management is added as separated functions

– Some functions are not included in the design:

– e.g. test capability for anomalies

– Some functions are not accessible for management:

Management System built externally

management:

– e.g. congestion control of transport layer

� Impact– Scalability, complexity, integration effort

– Self-management is not yet fully in place

– Increase of automation is still an open requirement

© 4WARD Consortium Confidential

Managed System(optical networks, radio access, etc.)

NM integrated throughexternal functions

04/03/2010 WP4/Slide 2

Approach for In-Network Management

� INM is design of co-located management functions

– Management tasks are delegated to the network elements• focus on architecture and software design

– Algorithms are distributed and autonomous (to high degree)

• focus on analysis of properties distributed systems

� Expected benefits:– Self-management to reduce capital and operational expenses

– Embedded functions to reduce integration effort

© 4WARD Consortium Confidential04/03/2010 WP4/Slide 3

Architecture for co-located management functions

� INM Framework is the enabler of INM functions– Three main elements: self-managing entities,

management capabilities, management domains– Enables construction of complex management

operations

� Self-managing Entities (SEs) encapsulate a network function with self-management functions– E.g. Generic Path with QoS monitoring

Global Management Point (GMP)

Operator

Service Access Point (SAP)

User

management

Ma

na

ge

me

nt

Inte

rfa

ce

– E.g. Generic Path with QoS monitoring– Defines high level interfaces– Enforces low-level objectives (service level)– Tells implementer to couple network function

with INM function (guidelines on next slide)

� Management Capabilities (MCs) implement the actual INM algorithms– Mapping between interfaces is co-designed– Different implementation approaches in Task 4.5

Co-location of management capabilities brings more efficiency (performance and integration effort)

© 4WARD Consortium Confidential

Management Domain

Objective

Enforcement

(downstream)

Objective

Monitoring

(upstream)

&

Manage-ment

CapabilityService

Logic

collocation

management

by objective

collab-oration managing

Self-Managing Entity (SE)

Manage-ment

Capability

co-design

Ma

na

ge

me

nt

Inte

rfa

ce

Service Interface

04/03/2010 WP4/Slide 4

Real-time management in the network

� Situation awareness consists in estimating the state of the networked system – We provide a set of distributed algorithms that provide views of the network state in

real-time

� Key issues:– Performance: accuracy and response time– Controllability: performance against overhead in a predictable manner– Controllability: performance against overhead in a predictable manner– Scalability: increasing size of network configurations– Robustness: maintain functionality even under adverse operating conditions (e.g. node

failures, local overload conditions etc.)

� Key approaches:– Distributed aggregation of local measurements to produce global estimates– Local management capabilities for specific self-management use cases

© 4WARD Consortium Confidential04/03/2010 WP4/Slide 5

An example

�Real-time monitoring of aggregated information in support of self-management– QoS assurance, service attacks, triggers re-

configuration

�Design and comparison of algorithms– Which degree of distribution and autonomicity

– Tree-based and gossip-based (high degree of

Changes of aggregated information

(network dynamics)

– Tree-based and gossip-based (high degree of autonomicity)

�Evaluation results– Both more efficient than traditional

– Two types of intervals (normal/stress conditions)

– Cost to trade-off performance/traffic (wireless/wired technologies vs services)

– Requirements on network dynamics to be considered in design

© 4WARD Consortium Confidential

Gossip-based has non homogenous overhead

04/03/2010 WP4/Slide 6

Management interfaces

� Monitoring performed in distributed manner– Information of the network function

is accessed locally

� INM Objectives exposed– Control behaviour of monitoring

– Allows composition

Control of objectives: accuracy, timelines etc.Composition in domains

– Allows composition

– No need to access individual counters externally

– Less integration effort between management and managed functions

© 4WARD Consortium Confidential

INM Aggregrate Capability

collaboration interface{

Publish aggregrate_update_Events

Listen for aggregrate_update_Events

}

organization interface{

set_aggregrate()

get_aggregrate()

get_counters(agg) }

Aggregratelogic

self-adaptation

collaboration

objectiveretrievalINM

AggregrateCapability

Collaboration between entities

Mapping performed inside the entity

04/03/2010 WP4/Slide 7

Impact on standards

� Management traditionally seen as vertical integration– Standards addressed mainly definition of alarms and policies– Direct interfaces between network element are excluded– Very granular interfaces

� Requirements for In-Network Management– Distribution of algorithms needs horizontal interfaces– Distribution of algorithms needs horizontal interfaces– Timeline of information dissemination is a key challenge for the Future Internet– Management Information more aggregated values

� Possible future directions– Standardization of the module structure of INM

• An additional benefit of interoperability and performances– Definition of high level information model for management

© 4WARD Consortium Confidential