Policy-Driven DISTRIBUTED MANAGEMENT ARCHITECTURES. (Towards Open Distributed Management Architecture) Kazi Farooqui

Transcription

1 Policy-Driven DISTRIBUTED MANAGEMENT ARCHITECTURES (Towards Open Distributed Management Architecture) Integrating Management and Distributed Object Technology Kazi Farooqui Department Of Computer Science, University of Ottawa, Ottawa K1N 6N5, Canada. 1

2 1. Integrating Management Models and ODP Model Model for Distributed Management: Synergy of: 1. Existing Management Models + 2. Distributed System Architectures such as RM-ODP. WHAT ODP Concepts + Principles + Architectural Framework can be used to enhance the OSI Management Model towards a Distributed Management Model? DISTRIBUTED OBJECT TECHNOLOGY (ODP, CORBA, ANSAware) NETWORK MANAGEMENT MODELS (OSI Management, TMN) DISTRIBUTED MANAGEMENT MODELS (ODMA) Figure 1. Integration of Distributed Object Technology in Network Management Models 2

3 Why DISTRIBUTED OBJECT TECHNOLOGIES: Management Activity: Inherently DISTRIBUTED in nature. Management Applications: Advanced DISTRIBUTED PRO- CESSING APPLICATIONS. Management Platforms: Require numerous DISTRIBUTED PROCESSING functions (Transaction, Replication, Group, Event Distribution, Location Transparency, etc.) How does ODP Help MOTIVATION GENERIC: A Generic Architectural Framework for Distributed Systems Development (Concepts + Architectural Principles + Distributed Processing Functions). META-STANDARD: Applicable to Different Application Domains (Telecommunications, Network Management, Manufacturing, etc.) DISTRIBUTED OBJECT MODEL: A Technology-Independent Model for the Development of Distributed Object-Based Systems. STRUCTURING PARADIGM: Different Viewpoints provide clear separation of concerns and focus on specific management aspects. (De-coupling of Management Applications and Management Platform). 3

4 Limitations of the OSI Management Model: Primitive Point-to-Point Model. (Support for traditional Point-to-Point Management Activity). Tied to a specific Communication Platform (CMIS/CMIP). (Tight coupling between management operations and communications infrastructure). Towards Distributed Management Models... Object-Machines for Configurable Distributed Management Platforms. Hierarchical Management Models. De-coupled Management Applications and Management Platforms. Policy-Driven Management Models. Management Domains. 3. State of the Art... 4

5 5

6 ODP Generic Model Enterprise Model Information Model Computational Model Engineering Model Inheritance & Specialization (of Concepts, rules, structures) Management Architecture Management Enterprise Model Management Information Model Management Computational Model Management Engineering Model TECHNOLOGY ARTIFACTS OF Management-SERVICES & Management-DISTRIBUTED PLATFORMS Figure 2. ODP Modelling of Management Systems 6

7 4. To Start... FOCUS is on: Preliminary COMPUTATIONAL and ENGINEERING Models of OSI Management. (Start with Bits and Pieces of Computational and Engineering Modelling of Management). ODP Computational Model: A Framework for the Definition of Management Application as a Distributed Object World interacting by invoking Operations/Notifications at Interfaces in a distribution-transparent manner. Identification of: 1. Management Application Components. 2. Multiple Interfaces of Components. 3. Interactions between Interfaces (Operations, Notifications). 4. Environment Constraints on Management Components and their Interactions. ODP Engineering Model: A Framework for the Definition of an Object-Based Distributed Management Platform. Modular Platform Support for Management Applications and of their interactions: 1. Communications-Support Infrastructure. 2. Transparency Support Infrastructure. 3. Domain-Specific Support Infrastructure. 4. Support for Management Policies. 5. Dynamic and Flexible Management Infrastructure. 7

8 5. Basic Computational Modelling of OSI Management What does ODP Computational Model offer to Management Model: Basic Object Model. Interaction Model. Type Model. Binding Model. 8

9 BASIC OBJECT MODEL: Legend: : manager role interface : managed role interface : application-specific role Figure 3. Management Object management object A Management Object offers multiple interfaces in different roles. Roles are associated not with Objects but with their Interfaces. Management Object may act as Manager on one interface and as Managed Object on another interface. Supports the Modelling of Hierarchical Management Applications : 1. Hierarchy of managers managing a set of distributed managed objects. 2. Delegation of responsibility from a manager to a submanager. (Delegation of Management). 9

10 INTERACTION MODEL: Computational Interface Template Distribution-Transparent Interaction Management Management Object Object Figure 4. Objects Interact by Operation Invocations OR Notifications at their Interfaces. Management Application is an object-based distributed application. Application components interact by sending operation invocations (ODP Interrogations) or notifications (ODP Announcement) at their interfaces. Management Interfaces specify: 1. signature: management operations and notifications. 2. behavior: ordering of operations and notifications = Protocol between interacting management interfaces. 3. environment constraints: QoS, Distribution Transparency, etc. (Notifications be delivered within 5 msec.) (Notifications be delivered to a group of managers). (Operations to be delivered to all or none of managed interfaces). 10

11 mco mco mco mco mco mco mco mco mco mco manager agent MIB MO MO MO MO Legend Managing system. Manager Interface. Application Specific Interface. Managed Interface. Managed system. mco: management computational object MIB: Management Information Base Figure 5. Distributed Management Computational Model Distributed Management Computational Model is Backwards compatible with the existing Point-to-Point OSI Management Model: 1. Managing System: a. Management Object = single-interface object with manager role. 2. Managed System: a. Management Object = Agent Object with a managed role interface and a manager role interface. b. Managing System manager interface is bound to the managed interface of Agent Object. c. Agent offers a local manager interface bound to MO 11

12 TYPE MODEL: M-CO1 M-CO2 M-CO3 Figure 6. Type Model: Interactions based on interface type matching. Management Applications are complex applications which can be created out of available components. A repository of management components and their interface types enables faster application creation. Management Application creation/composition based upon type-checked (late) binding of (interfaces of) application components. substitutability of one application (component) by another. (A manager interface can be substituted by its sub-type). Current ODP Type Model: Based on operation signature compatibility. Subtyping can be extended to QoS compatibility and behavior compatibility to meet the needs of management applications. 12

13 BINDING MODEL: M-CO-1 Configuration Management Interface M-CO-2 BINDING OBJECT M-CO-3 QoS Management Interface M-CO-4 Figure 7. Multi-Manager / Multi-Managed Object Binding. Support for management application-specific interaction pattern between multiple managers and managed objects. Binding Objects enable complex binding structures to be supported between management application components. (Multicasts of Notifications issued by an Event Forwarding Discriminator to multiple managers). Applications have control on configuration and quality of service of the complex binding (Add / Delete Managed Objects from the configuration). 13

14 6. Basic Engineering Modelling of OSI Management What does ODP Engineering Model offer to Management Model: Model of an Object-Based Distributed Platform: 1. Concepts, Rules, Structuring Framework (of nodes, capsules, clusters, objects) for the organization of a Management Platform that supports execution and interaction of Management Application Components. Distribution Transparency Support Mechanisms: 1. An engineering framework of transparency support mechanisms for the provision of location transparency, migration transparency, transaction transparency, group transparency etc. to the interactions between Management Application components. Model of a Configurable and Policy-Driven Platform. 14

15 Basic Engineering Concepts of OSI Management: 1. Basic Engineering Object: Engineering representation of management computational object. m3 m4 mco m1 m5 m2 m2 m5 meo meo m1 m4 meo meo Figure 8. Computational Object and Cluster: A powerful modelling paradigm. 2. Cluster: A group of related management objects at a node. (A set of related managers synchronized via internal interfaces and accomplishing a common management function OR a set of MOs related to different aspects of a single resource). 3. Cluster Manager: Part of Agent Functionality. 4. Capsule: A group of management clusters with unrelated management responsibilities. 5. Capsule Manager: A Super-Agent. 6. Nucleus: Access to CMIP communication facility. m3 Legend: mco: management computational object meo: management engineering object 15

16 7. Node: An open system consisting of both manager and managed interfaces. 8. Stub Object: A marshalling/information modification functionality that transforms management operations into messages that are exchanged as CMIP PDUs. 9. Binder Object: Support for Distribution Transparencies. 10. Protocol Object: CMIP and the underlying stack of OSI/TCP- IP communication protocols. 11. Channel: A configuration of stubs, binders and protocol objects between manager and managed interfaces is the key to the design of a policy-driven management platform. 16

17 Extending the current OSI Management Model: Engineering Functions clubbed in the Agent: 1. Scoping (performed in a local system) 2. Filtering (performed in a local system) 3. Synchronization (performed in local system). 4. Event Notification 5. Object Management. Scope of these functions can be extended in a distributed management context: 1. Scoping -----> ODP Group Function. 2. Synchronization -----> ODP Transaction Function 3. Event Notification -----> ODP Event Notification Function 4. Object Management -----> Cluster Manager/Capsule Manager 17

18 SCOPING: Scoping can be performed on Managed Objects (MIBs) in multiple systems using the ODP Group concept. ODP Group MIB Managed System-1 MIB Managed System-2 Scoping MIB Managed System-3 Figure 9. Grouping and Scoping in a Distributed Management System. MIBs on distributed systems are organized as an ODP Group with Base-Level MOs as members. A management operation multicast by a manager is received by all group members. Group Invocation: Addressing and selection of base-level MOs in different systems. Scoping: Selection of MOs in the Local System. 18

19 SYNCHRONIZATION: In a distributed management model, manager requires a management operation performed atomically on all MOs distributed on different open systems. Synchronization can be performed on Managed Objects (MIBs) in multiple systems using the ODP Transaction concept. Manager Agent MO MO MO MO TF TF (M) To other (S) Managed Systems CMIP CMIP OSI Protocol Stack OSI Protocol Stack LSF Legend: TF: Transaction Function M/S:Master / Subordinate LSF: Local Synchronization Function Figure 10. Transaction in a Distributed Management System Management operations emitted by the manager are intercepted by the Transaction Function. Transaction Protocol is carried over CMIP using M-Action PDUs between the involved manager and managed systems. Local Synchronization is performed by LSF. 19

20 EVENT NOTIFICATION & OBJECT MANAGEMENT: Manager Agent MO MO MO MO Legend: CPM: Capsule Manager CLM: Cluster Manager ENF: Event Notification Function CMIP OSI Protocol Stack Event Notification CMIP OSI Protocol Stack CPM/ CLM Object Management To other Clusters ENF Event Notification Event Logging To other Clusters Figure 11. Event Notification and Object Management Function in the ODP Framework Both the MOs (in different systems) and the Manager subscribe to the ODP Event Notification Function to send and receive events. ODP Cluster/Capsule Manager perform the object management functions such as MO creation, deletion, checkpointing, deactivation and reactivation. IN ALL THESE CASES AGENT WORKS AS AN INTER- CEPTOR OF MANAGEMENT OPERATIONS. 20

21 7. Policy-Driven Distributed Management Models Basic Components 1. Monitoring: To obtain information from management interfaces of physical resources 2. Domains: To group management objects and to partition responsibility. 3. Policy: To permit the behavior of automated managers to be modified without re-implementation. Management Domain A set of objects which are related because a certain part of their behavior is controlled by the same authority (e.g. telecom stakeholder) under a set of management policies. - Autonomy (Authority) - Policy - Behavior/Aspect - Addressing Unit X-Aspect Management Domain, Y-Aspect Management Domain Start_Test() on all domain members for which the predicate xattribute > n is true. 21

22 Why Domains 1. To group the objects related to a particular service or function. 2. To group the objects to which a policy applies. 3. To specify boundaries of management responsibility or authority. 4. To provide a naming context in which managers may assign local names of their choice to the objects they manage. 5. To reflect organizational, geographic or network structure. Domain Membership 1. Create a domain 2. Delete a domain 3. Add an object 4. Delete an object 5. List members 6. Change member name 7. Query domain s parents. 22

23 Policy-Driven Management 1. Mechanisms vs. Policies: Separation of management policy from Automated Managers which interpret the policies. 2. Specification of Policies: High-level Specification of management goals using some Policy Specification Notations. 3. Interpretation of Polices: Interpretation of policies into management actions. policy object PAI MPI interpret manager object operations notifications replies Legend: : Object Management Interface : Manager Interface : Managed Interface managed object MPI physical resource application specific interfaces PRI: Physical Resource Interface MPI: Management Policy Interface PAI: Policy Administration Interface Figure 12. Policy-Driven Management: Managers interpret Policy to perform management PRI 23

24 Managed Domain-D MMI PMI DMI Domain Administration Authority Policy Object sub-domain D-1 MMI PMI DMO MO- 1 DMI MO- 2 MO- 3 Manager Object Manager Domain OPR REP/NTF sub-domain D-2 MMI PMI DMO MO- 4 DMI Legend: OPR: Operation Invocation NTF: Notification Invocation REP: Reply Invocation MMI: Membership Management Interface PMI: Policy Management Interface DMI: Domain Management Interface : Object Management Interface : Managed Interface : Manager Interface MO- 5 MO- 6 MO- 7 Figure 13: Computational Model-1: Policy-Driven Managers 24

25 Managed Domain-D sub-domain D-1 OPR NTF/REP MMI MMI PMI DMO PMI DMO MO- 1 DMI DMI Policy (P) Manager Domain MANAGER OBJECT OPR NTF/REP OPR NTF/REP OPR NTF/REP OPR MO- 2 MO- 3 MO- 4 NTF/REP Policy Conflict Resolver OPR MO- 5 NTF/REP Legend: OPR: Operation Invocation NTF: Notification Invocation REP: Reply Invocation OPR MMI: Membership Management MO- 6 NTF/REP Interface PMI: Policy Management Interface DMI: Domain Management OPR Interface MO- 7 NTF/REP : Object Management Interface : Managed Interface : Manager Interface Figure 14. Computational Model-2: Policy-Driven Managed Objects sub-domain D-2 MMI PMI DMO DMI Policy Conflict Resolver Policy (P1) Policy (P2) 25

26 Policy (P1) Policy (P3) Manager Domain M-1 G S M C M I P C M IP G S M MO-1 PCR M-2 M-3 sub-domain M-4 G S M G S M G S M C M I P C M I P C M I P Group Communication using Multicasting Protocols Engineering Support for Domains C M IP C M IP C M IP G S M G S M G S M MO-2 sub-domain MO-3 MO-4 Managed Domain PCR Policy (P2) Legend: M: Manager Object MO: Managed Object PCR: Policy Conflict Resolver Object GSM: Group Support Machine Policy (P4) Figure 15. Policy Driven Manager and Managed Objects 26

27 ODP Distribution Transparencies in the Management Model: Access Transparency: Management Applications invoke CMIS service through standardized service primitives. Location Transparency: Requires location-independent naming model in the management domain. Migration Transparency: Highly dynamic mobile telecommunications environments. Transaction Transparency: Management Operations (such as Suspend_Test(), Resume_Test(), Set_Attributes()) require all-or-nothing semantics in their execution. Group Transparency: Identical management operations on a set of distributed Managed Objects. Persistence Transparency: Reactivate both manager and managed objects when a notification or an operation is invoked on them. 27

28 Management constitute a significant application domain of ODP. ODP Provides: Distributed Object Model. Modeling Frameworks (viewpoints). Interoperability Model. 7. CONCLUSION Distribution Transparency Model. Increasing convergence taking place between management models and distributed computing (such as ODMA). ODMA: Object-Based Management Architecture Distributed Management Model Technology-Independent Management Policy-Driven Platform Configurable Platform. Distributed-Object Computing Technologies have a major role to play in the definition of tomorrows MANAGEMENT ARCHITECTURES. 28