Moving Knowledge Agents for Network and Service Management Tianning Zhang GMD FOKUS Research Institute for Open Communication Systems German National Research Center for Information Technology Hardenbergplatz 2 10623 Berlin Germany e-mail:
[email protected] fax: + 49 30 25499 202 Abstract The emerging global information infrastructure requires new framework for moving network and service management solutions from producers to consumers. Traditional mechanisms for such movement require sophisticated human intelligence and expertise, high costs in installation and maintenance, or impose high traffic load/availability requirement on the global networks. This paper proposes a new framework for developing, marketing and deploying management solutions using Java-based Mobile Intelligent Agents. The management solutions are regarded as knowledge rules, defined as Java objects and structured into a hierarchy by the hierarchical specialization relationship among the problem situations which the solutions are addressing. With this hierarchical organization of solutions, we can obtain the Object-Oriented style of reusability/interoperability, and accumulative construction/ standardization, of management functionalities in the distributed open environment. To manage the real resources, Java-based Mobile Intelligent Agents (IA) carrying a hierarchy of management solutions are envisaged, which travel among the managed network sites to solve the problems in the environment, by finding out and applying locally the most suitable management solution. Key-words:
Intelligent Agents, Mobile Agents, Network and Service Management, TMN, Java, Knowledge/rule-based Systems
Submission area:
AI techniques for network and services management (Intelligent Agents, Expert Systems, etc...)
1
1
Background
The heterogeneity, distribution, scale, and dynamic nature of emerging global Infotmation Infrastructure ([2] [12]) is calling for new paragigms for the development, deployment and management of the open resources. Most resources in this environment will be provided to the consumers through a long value added chain of producers, providers, vendors etc., which are widely separated by administrative or geographical distances. The complicated and versatile inter-relationships/distributions of the resource-associated players and roles have significant impact on the frameworks for managing the resources. In this context, management solutions addressing problems related to the resources can be developed by any players in any operation or role contexts, and in any geographical or administrative positions within the open world of telecommunication. At the same time, in a global market, each of such management solutions will be used in a variety of heterogeneous domains, which can be geographically or administratively far away from the source or provider of the solution. The effectiveness and the market chance of these management solutions depend heavily on the following characteristics: – Intelligence Once a management solution leaves its manufacturer, it will be installed at some remote customer sites and has to operate autonomously in a foreign environment. Such reusability of the management solution depends on its ability, or intelligence, to cope with all the complicated situations that can occur in the new environment. – Mobility As the management solutions have to be installed at (or delegated to) different sites, mobility in the sense of platform independence and the support for porting among sites will be a key issue. In a traditional open system management scenario, both intelligence and mobility of the management solutions are realized by human experts from provider or customer organizations, who will: •
carry the management solutions in some media, like disk, CD, tape etc., travel to the customer sites, install there the solutions and assist the customer in solving the problems with the solutions
•
download the management solutions, together with the menus and user guides, from some sources, install and use the solutions in the local environment by following the menus and guides
•
choose the management solutions based on the information collected from the customer sites, and call the corresponding management functions directly via remote invocations, like Remote Procedure Call (RPC), RMI[14], or SNMP/CMIP [7][8] operations.
The major problem here is the enormous amount of expertise needed for coping with the versatility and complexity of the problems in the networks. With the increasing complexity of telecommunications environments, lack of enough qualified personnel is becoming nowadays a major obstacle in implementing effective management. Management based on downloading, which is increasingly popular due to its association to Internet and WWW, can greatly reduce the cost and delay associated to the travelling experts. However, it still imposes high demand on the expertise of the customers to install and operate the downloaded solutions. Management based on remote invocations requires the pre-installation of the management solution software at the customer sites. This kind of pre-installation can take up a significant amount of local resources due to the number of possible management solutions that can be relevant for the local environment. Moreover, the remote interactions between server/client can impose excessive burden on the communication network and introduce excessive delays in the management procedures.
2
It is in this context where the concept of Mobile Intelligent Agents (henceforth called IA for simplicity) enters the arena and obtains its importance. The concept of IA originates from the area of artificial intelligence [1]. Recently IA is becoming the fashion word in a large community of computer scientists with very heterogeneous but co-related interest areas like software engineering, human interfaces, network communication, knowledge-based systems, and even psychology. Generally speaking, a Mobile Intelligent Agent can be looked upon as an autonomous entity, which can migrate to different nodes of the telecommunication environment and which, with certain degree of “intelligence”, takes up there part of the tasks of human experts. The most important difference of IA from other current approaches is its ability to install itself and operate autonomously in a new environment. In this way, an IA-based solution for open system management supports the following advantages and possibilities: •
reducing the number of remote interactions and the associated traffic load for realizing the management functionality, via autonomous IA operations
•
reducing the requirement on human intelligence or personnel qualifications during the installation and operation of the management solution, by utilizing the IA intelligence
•
enabling a global open service market for network and service management solutions, where a management solution can be produced, traded, transported and consumed.
This paper discusses a new IA framework for supporting network and service management in a global, open telecommunications environment. This framework is currently under validation in GMD FOKUS.
2
Pre-requisites for an IA-based Solution
Following the above definition of IA, an IA framework for network and service management solutions should mainly support of two functions: 2.1
IA Knowledge Representation
IA knowledge, which in fact defines the IA management functionality, is needed for supporting the autonomous management of resources. In an open environment, numerous players can contribute to the management solutions with there own experiences and views. A suitable framework supporting such a global and co-operative acquisition of knowledge will therefore determine the success or failure of the IA-based approach. The following characteristics of such a knowledge acquisition framework will play a key role: – Reusability and Interoperability The global telecommunication environment encompasses a large number of heterogeneous sub-environments. The success of any solutions in this world depends on their capability of integrating themselves into the operation environment, i.e. the ability to be applied in a wide range of contexts. Such an ability has to be realized in one of the two ways: •
to use the same knowledge or functionality in constructing a broad range of systems and applications, i.e. to have reusable components
•
to use the same knowledge or functionality in a variety of environments, especially heterogeneous environments. I.e. to have a function that can operate in a broader range of environments.
Being important in all the scientific areas, reusability and interoperability provide the dominating requirements
3
in the case of distributed network resource management. Reusability and interoperability in the distributed environment are usually achieved via standardization at different syntactical or semantic levels, and through Object-Oriented constructions. – Accumulative Construction The complexity of the network and service management scenarios has to be coped with by using an enormous amount of very complicated knowledge. Such knowledge can only be derived from profound experiences which have to be collected gradually during the evolution the resources. Management has to be done during the whole life cycle of the resources. The result is that it has to, at the beginning, be based on some very primitive management knowledge and operations. With the time passes by, new knowledge will be obtained concerning the deeper inter-relationships between the components of the environment and the detailed behaviour of the components. New management operations can also be added due to new requirements and new facilities. All these new knowledge and functionalities have to be gradually integrated into the management applications. In this sense, accumulative construction of the intelligent management systems is also an important factor in realizing effective management of network resources. The most important technology in this context is the ObjectOriented style of development. 2.2
IA Mobility IA mobility is supported by two sub-functions: •
the IA transportation
which supports the migration of the IAs among different network nodes, and •
the IA execution
which supports the execution of IA functionality within a network node. Transportation or migration of IAs have to be realized via coding the IA knowledge and functionality in the source node, transporting the coded object across the network, and installing it at the destination network node after decoding the object. The necessary and sufficient pre-condition in this case is that all the network nodes support the same IA knowledge representation and communication protocol, either directly or via some proxy functions. A representation language which is supported by most popular platforms will certainly have advantage over other less popular languages. With the dominance of Internet in the context of telecommunication and telecommunications management, any language platforms that are widely accepted and supported in this community will be the best candidate. Java, due to its strong connection to WWW, and also due to its simplicity in Internet-oriented programming, is currently very popular within the Internet community and is also supported by most platforms. Based on this consideration, we select the Java as the base language for our IA-based management solutions. The major restriction with Java is that it lacks some direct support for agent programming provided by some other languages like Telescript [6], which are designed almost solely for supporting the development of mobile agents. E.g. the most outstanding advantage of Telescript over Java is the direct continuation via the go operation, which means that an IA can make some operations in one network node, move to another node and simply continue with rest of the operations there. In case of Java, such continuation have to be supported by state information and switch statements on such state information. In our framework, we expect state-based continuation to be a more suitable solution due to the clear knowledge inference phases of the IAs. We also expect that Java will support all features needed.
4
Transportation of IAs over the network can be supported by different high level protocols or services. In our approach, the development of IA framework is based on the serialization/deserialization [15] mechanism from Java. This Java mechanism is used to produced a serialized (coded) IA object which can be transmitted over the network by conventional socket communication facility, and to deserialize the object at the destination to recover the IA. The responsibility of the IA execution function is to take over the IA migrated to this node, and to realize the functionalities specified in the IA by interpreting the knowledge contained. In this context, the IA execution environment has to solve the following two problems: Generic Platform Except for the restrictions imposed by the IA framework, the IA execution environment has no knowledge about the specific functionality of the IA that will enter its domain. Therefore the IA execution architecture has to based on very generic mechanisms that interpret the IA knowledge to realized the required functionalities. Rulebased architecture, which is very popular in artificial intelligence applications, can play an important role in this context. •
Co-operation
As each IA operates in a foreign, distributed environment, and carries only very limited knowledge provided by its factory and collected from the network nodes it has visited, it has to co-operate with different entities, both local and remote, to fulfil its tasks. On the one hand, the IA usually has to contact its source for reporting the progress of its operation or to get more consultation from the source. In some case it is also necessary for one IA to co-operate with other living IAs to realize a global management objective. On the other hand, the IA has to co-operate with the local management entities, especially the management agents, to collect local information and to test/manipulate the local resources. Sometimes it is also necessary for an IA to extend, or partially replace its knowledge with the local proprietary knowledge to enable an optimal management of the local resources. Such co-operations have to be supported by some information exchange protocols, software APIs and by a flexible knowledge representation. Other pre-requisites, which is also critical to an IA framework for network management, e.g. security aspects of IA mobility and executions, are not yet addressed by the research presented in this paper.
3
The IA Knowledge Representation
The knowledge for the management IA is the set of management solutions that address the problems that can occur in the network. Each such management solution can be constructed from two key components:
3.1
•
the test function for the pre-condition of the solution, which is defined as the network situations that require or support the application/invocation of the solution, and
•
the management plan, defined as a pre-defined sequence of management actions, including the generation of Event Reports, that can lead to the solution of the problems in the network. The Situation Theory for IA-based Network Management
Most currently dominating network and service management frameworks, including the ITU-T TMN[8]or Internet SMI [7], view the managed network environment with its set of resources as a set of Managed Objects (MOs), as depicted in figure 1.
5
Figure 1: Network and Service Management View ls co to ro tP en em ag t an en M ag
manager
managed environment/
Managed Objects
In this context, the managed resources like switches, hosts and processes are controlled by some management agents. Such management agents implement the abstract view of the underlying resources, defined as the (MOs). Such MOs can be managed or processed by the managing stations or managing processes via standardized management protocols. The protocol, which is SNMP in case of SMI and CMIP in case of TMN, and the specification of the managed information, which is based on either the SMI/SNMP specification methodology or the TMN/ GDMO [10] methodology, determine together the execution environments for any network managing entities including the IAs. In this paper, we discuss only the TMN/CMIP-based management environment. The framework, however, can be easily adapted to a SNMP-based environment. In a TMN/CMIP-based environment, the MOs are defined by their GDMO specifications. Each such MO can possess the following information capabilities: •
the attributes, which contain the status and characteristics information concerning the managed resources,
•
management actions, which support the management operations upon the underlying resources, and
•
the notifications with their parameters, which the managed resources can issue to the managers for signifying special events in the network. In the following we will call notifications received by the managers Event Reports (ERs) to maintain a simple terminology.
A manager in this context can make the following basic operations on the MOs: •
get for obtaining the value of an attribute by proving the Identifier of the MO and the attribute
•
set for setting the value of a MO attribute
•
call a management action which is supported by the MO
•
receive an ER from the agent
In such an environment, the current situation in the network is characterized by the set of MO instances (with their values) and the ERs (with their parameters) that have been generated. If we denote the set of all possible sets of MO instances (with all their values) as Ψ and the set of all possible sets of generated ERs as Φ, we can have the following definitions ([16]): DEFINITION: A situation is the union of an element from Ψ and an element from Φ. I.e. the universe of situations is Ε = {λ ∪ α | where λ ∈ Ψ and α ∈ Φ}.
6
DEFINITION: A situation concept ε is a subset of Ε, i.e. ε⊆Ε With these definitions, a situation concept characterizes some possible situations that belong to a specific category and can be used to specify the pre-condition for management solutions. Based on the definitions, it is not difficult to derive a hierarchical structuring of the situation concepts using the specialization/generalization relationships. Definition: A situation concept ε is a specialization of another situation concept σ, or σ is a generalization of ε, iff ε ⊆ σ. From the basic properties of set inclusion, it can be derived that the specialization relationship on situation concepts is transitive and anti-symmetric, i.e. it imposes a hierarchical structuring on the situation concepts. Theorem: For arbitrary situation concept ε, σ, ε ∩ σ is always the specialization of ε, where ∩ is the normal set conjunction operation. To test and determine the current situation in the network, we can apply situation test, which checks whether the current situation in the next work corresponds to a specific situation concept. Definition: A situation test τ is a function (i.e. an operation) from Ε to true or false, i.e. τ : Ε → { true, false} The denotation (semantics) of a situation test τ is the situation concept ε, such that an element χ is in ε iff we have: τ(χ) → true I.e. a situation test corresponds to a situation concept and can be used to the check to validity of a situation concept in the network. Following to the hierarchical structuring of situation concepts, we can derive the hierarchical structure on the situation tests. Definition: A situation test τ is a specialization of another situation test γ iff the denotation for τ is a specialization of the denotation for γ. If we have a side-effect free conjunction, denoted as “;”, of two situation tests, we can also have the following result: Theorem: The side-effect free conjunction of the a situation test τ and any other situation tests is a specialization of τ, and is independent of the order of the tests. Intuitively, a specialized situation concept contains fewer real situations and can make a closer approximation of the situation in the network. Correspondingly, a specialized situation test contains more restrictions on its denotation concept and can therefore be satisfied by a smaller set of network situations (see [16] [17]). Management plans based on specialized situation tests can therefore support safer, and more effective solutions of the current problems in the network. In our Java-based management environment, the pre-condition for a management solutions is specified by the situation tests that characterize the possible network situations for the management actions (management plan).
7
To guarantee the side-effect free testing of the managed network environment, we restrict the CMIP management operations within a situation test to the CMIP get. With some approximation, we can consider the tests as side-effect free. The result of the multiple tests will be independent of the order in which individual test is carried out, and is always the specialization of a single test. A situation test in our IA framework is just a boolean Java method which uses only the get methods from the CMIP API and any other Java basic methods. This property will be guaranteed by the editing environment for defining the Management Solutions. 3.2
The IA Management Solutions
As discussed above, a management solution consists of a situation test, which checks the situation in the environment, and a management plan for addressing the corresponding problems in the network by issuing management operations. For this purpose, a management solution will be a subclass of the following abstract class definition: public abstract class ManagementSolution { public EventMemory eventMemory; public Event currentEvent; public CmipAgent agentAddress; public abstract boolean situationTest (); public abstract AgentHost managementPlan (); } In this definition, the EventMemory class is used to keep a list of ERs, each of which belonging to the class Event. The EventMemory class also supports the manipulation of events stored via operations like filtering, searching, deleting events, adding ERs, refreshing the memory etc. The currentEvent variable is used to hold the ER that currently drives the processing of the IA, while the eventMemory maintains the list of ERs that can be relevant in the environment of the IA. The agentAddress serves as the contact address for all the CMIP operations that are to be carried out within the situation test and management plan methods. The result of the managementPlan can be either null, in this case the IA will stay in the local environment, or it can be an AgentHost -- the next place to which the IA should be moved. A real management solution, which extends this abstract class, can also contain other private variables that can be manipulated by the situation test method to carry extra test result information from the test phase to the management plan. Based on the hierarchical structuring of the situation tests, we have also the hierarchical relationship among the management solutions. For building up such a hierarchy of management solutions, we impose the following conventions on the development of management solutions: •
Each management solution will be contained in a separate Java package. Each management solution package for a more specific problem is positioned under the package for a more generic solution.
•
The situation test within each management solution implies automatically the situation tests in all the ancestor (more generic) management solutions.
8
Following the Java convention, the management solutions are stored in a hierarchy of directories in the file system as depicted in figure 2. E.g. a management solution for addressing some general problems of ATM switches can be specified in the package “...atmSwitch”, and stored in the directory “.../atmSwitch/”. A new management solution for more specific domain, like the Fore ATM switch, can be specified in a child package “...atmSwitch.fore”, and will be saved in the directory “.../atmSwitch/fore/”. Figure 2: The Hierarchy of Management Solutions package and Management Solution name
atmSwitch/
atmSwitch.Solution synoptics/
fore/ atmSwitch.fore.Solution vpFault/
directory name
portAlarm/
atmSwitch.synoptics. Solution
atmSwitch.fore.vpFault. atmSwitch.fore.portAlarm. Solution Solution
The situation test in a child node in this hierarchy, e.g. for the management solution atmSwitch.fore.vpFault, automatically implies that the tests in the ancestor nodes, atmSwitch and atmSwitch.fore in this case, must be also satisfied. In this way we build up a hierarchy of management solutions which starts with generic management solutions at the upper part, and grows with specific management solutions positioned under more generic solutions. The management solutions will be pre-compiled to the Java class files. At the time of IA creation, these class files will be loaded to build a corresponding hierarchical Java data structure, called the managementHierarchy, which contains the management solution classes.
4
IA Mobility and Execution An IA consists of the following key components: •
the eventMemory for holding the set of events collected by the IA during its life time, and which are still relevant for the IA processing
•
the currentEvent which is driving the inferencing of the current IA
•
the managementHierachy of management solutions
The IA generation, execution and transportation are supported via two Java Threads, as depicted in figure 3. In this configuration: •
the IA Factory is responsible for receiving ERs from the managed resources via a Java/CMIP proxy and the TMN Event Forwarding [9] service. Upon, receiving such an ER, the IA Factory generates an IA with empty managementHierarchy, and only with one ER in the eventMemory referring to the currentEvent. This initial IA will be then sent to the associated IA Host for this IA Factory.
9
Figure 3: IA Execution/Transportation Environment
MOs/CMIP Agent
ER
IA Factory
ER Java/CMIP CMIP actions/ notifications
Proxy
IA delegation
managementHierarchy
Java IA Host action/notification invokations
Loading management solutions
delegation of IA to other IA Hosts
•
the IA Host is responsible for inferencing based on the knowledge in managementHierarchy, and for finding out the most suitable management solutions for the current situation in the network. For this purpose, the IA Host also maintains an eventMemory of the ERs received in the local environment.
It is possible that some network nodes have only IA Factory, which means that the initial IAs will be sent to some associated IA Hosts in remote network nodes. Each time when an IA enters an IA Host, the IA Host first integrates the local management solutions hierarchy into the IA, and then starts inferencing on knowledge hierarchy to find the suitable management operations to be issued to the managed resources via the Proxy. To update the management solution hierarchy, the IA Host traverses through the local hierarchy of the Java management solution package file directories, creates an record of each management solution class, and builds up a hierarchy of management solutions (implemented in Java as vector of vectors) to be assigned to the corresponding IA variable. The IA Host will also merge the eventMemory of the IA Host to the eventMemory of the IA After the creation of the managementHierarchy and the initialization of other variables, the IA Host starts the execution of the IA, i.e. the traversing of the hierarchy to find a management solution for the current situation in the network. The traversing will be done in pre-order and started at the root of the hierachy. At each management solution node the IA Host will do the following: 1.
create and initialize the management solution instance
2.
call the situationTest method
3.
if the result of situationTest is true, call the managementPlan method in that solution, otherwise continue the traversing of the hierachy on the next management solution node
The result of the managementPlan method has three possibility: 1. the method returns null, and sets the currentEvent variable to null In this case the IA will be terminated. 2. the method returns null, but sets the currentEvent variable to a new ER (one from the eventMemeory, or a newly generated one) In this case the IA Host will set the hierarchy to the subhierarchy under the current node and restart the process
10
of traversing, based on the intermediate result coded in the new ER. 3. the method returns a new agentHost address different from the current host In this case, the IA host sets the managementHierarchy to the subhierarchy under the current node and moves the whole IA instance to the new IA Host via the Java Serialization mechanism (figure 4). In the new network node, the new IA Host first updates the eventMemory by adding to it its local events, and then tries to update the hierachy of management solutions with its local hierachy of solutions. To update the management hierachy, the new IA Host can also replace those nodes in the IA which are also defined in the local hierarchy (with the same package names) and the descendant nodes of such shared nodes. After updating the IA, the IA host starts the IA execution just like procedure described above. The advantage with this whole scheme is that each local network node can maintain its proprietary solutions for its local resources. The travelling IA will consider these local solutions even if these are not available in the origin node of the IA. Figure 4: IA Execution and Mobility eventMemory currentEvent feedback the new management solution hierachy
Serialization IA move pre-order searching of the hierarchy
In the new IA Host
5
In the precedent IA Host
Conclusion
This paper proposes a new framework, with the flavour of Object-Oriented development, for the Java IA-based management of telecommunication network resources. The framework is based on the collection of management solutions in an open global network and service environment, the organization such management solutions into some global knowledge hierarchies following the application context situations of such solutions, and on using such hierarchies to support the mobile IA in finding out an optimal solution for the problems existing in the networks. The whole approach has, among others, the following major advantages: •
Automation and delegation of management processes via mobile, knowledge-based Agents
The management framework enables the delegation of autonomous management solutions. In this way, an enterprise can Outsource ([3], [4]) part of the management responsibility and management problems to some specialized management solution providers. Such outsourcing and the automation of management processes can greatly reduce the complexity and costs within the enterprise for managing the enterprise network resources. With the increasing complexity of broadband telecommunication environments, such reduction of costs will be a major issue in applying any management frameworks.
11
•
Compatibility with existing environments
The approach is designed to be based on the TMN management framework and can easily be adapted to a SNMP environment. Each IA communicates with the managed resources via standardized management protocols and operates on the management view offered by the Management Information Base (MIB). The co-operation among the IAs is realized via TMN Event Reports and Event Forwarding [9] service, instead of novel co-operation languages like KQLM [5]. This approach can be therefore easily embedded in a standard conform network and service management environment. •
Reusability and Interoperability
A management solution hierarchy starts with some generic knowledge/solutions which are applicable in different environments, and it derives specific knowledge by adding specific information about the individual environment classes. If applied in a different environment, the generic management knowledge in the upper part of the hierarchy will continue to be valid, while the lower part situation knowledge will not be satisfied and will be simply ignored. The result is that the IA with this hierachy operates also in a new environment, although with some loss of effectiveness. For the same reason, one management solution hierachy can be reused in different environments to build there the management systems. •
Accumulative and distributed construction of the global knowledge
When starting to manage an environment with an IA-based framework, we usually use a partial management solution hierarchy to initialize the management functionality. Later, with experiences and expertise, we can added more specific management solutions with new situation tests and their associated management plans, to improve the performance of the IA system. Moreover, the management solutions can be developed/advertised by any players in anywhere of the open world. The players can also decide to maintain such solutions only in the proprietary environments, and to ask IAs who visit their sites to use such proprietary solutions instead of the generic ones. •
Standardization of global available management solutions
The hierarchical structuring of the management solutions in the Object-Oriented style enables the standardization of the relatively generic management solutions in various contexts and domains for the upper part of the hierarchy under the root. Such management solutions can be further made available to all users via hierarchical information services like the ITU-T X.500 Directory Services[11]. With these characteristics and possibilities, we expect the approach to provide a realistic management framework for the current and emerging telecommunication networks and services.
6 [1] [2] [3] [4] [5] [6] [7]
References Communication of the ACM Journal, Intelligent Agent, Vol.37, No. 7, July 1994 ETSI, SRC6 Final Report on European Information Infrastructure, June 1995. F. Dotti, S. Covaci, ODP Viewpoints of Management Outsourcing, Proceedings of the IEEE /IFIP Network and Management Symposium (NOMS’96), Kyoto, Japan, April 1996 F. Dotti, Management Outsourcing in Open Distributed Environments, Proceedings of the Enterprise Networking Workshop (ENW-96), Dallas, June 1996 T. Finin, KQLM as a Agent Communication Language, Proceedings of the 3rd International Conference on Information and Knowledge Management (CIKM’94), Amsterdam, 1994. General Magic, Telescript Technology: Mobile Agents, http://www.genmagic.com/Telescript/Whitepapersh/wp4/whitepaper-4.html, 1996 IETF, Structure and Identification of Management Information for TCP/IP-based Internets, RFC 1155,
12
[8] [9] [10] [11] [12] [13] [14] [15] [16] [17]
May 1990. ITU-T, Recommendation M.3010, Principles for a Telecommunications Management Network, March 1995. ITU-T, Recommendation X.724, Information Technology - Open Systems Interconnection - System Management: Event Report Management Function, 1993 ITU-T, Recommendation X.722, Information technology - Open System Interconnection - Structure of Management Information: Guidelines for the Definition of Managed Objects, January 1992. ITU-T, Recommendation X.500, Information technology - Open System Interconnection - The directory: Overview of concepts, models, and services, November 1995 NIST, United States National Information Infrastructure Virtual Library, http://nii.nist.gov/nii.html, 14 October 1996 OMG, The Common Object Request Broker: Architecture and Specification, Draft 29, December 1993. Sun Microsystem, Java(TM) Remote Method Invocation Specification, 1996 Sun Microsystems, Java Object Serialization Specification, May 1996 T. Zhang, S. Covaci, The Semantics of Network Management Information, Proceedings of the 1996 IEEE INFOCOM Conference, San Francisco, March 1996. T. Zhang, S. Covaci and Radu Popescu-Zeletin, Intelligent Agents in Network Management, Proceedings of the 1996 IEEE GLOBECOM Conference, London, 18-22 November 1996.
13
