Future Architecture and Mechanisms of the Self-Managing Internet

This work was funded by EFIPSANS under Grant No. INFSO-ICT-215549, National Basic Research Program of China (“973” Program) under Grant No. 2009CB320504, National Natural Science Foundation of China under Grant No. 60672086, and Sino-Swedish Strategic Cooperative Program on Next Generation Networks under Grant No. 2008DFA12110.

    In current Internet research, the design objective, architecture, implementation and features of the future Internet are being discussed, and experts generally hold different views[1-5]. Experts within the European Union Seventh Framework Programme (FP7) EFIPSANS (Exposing the Features in IP version Six protocols that can be exploited/extended for the purposes of designing/building Autonomic Networks and Services) believe self-management will be one of the main features of the future Internet. The service flexibility, network reliability and availability of a network can be markedly improved by self-management, and the operational and maintenance costs can be significantly reduced.

    The main purpose of EFIPSANS is to propose a new architecture for the Internet, describing its behavioral features and how they might be implemented. A number of research institutes, universities and companies participate in the project, including the Fraunhofer FOKUS in Germany, the Waterford Institute of Technology (TSSG) in Ireland, the National Technical University of Athens, the Warsaw University of Technology, the University of Luxembourg, Ericsson in Sweden, Alcatel-Lucent in France, and Telefónica in Spain. The Beijing University of Posts and Telecommunications (BUPT) is the only non-EU member of project.

1 Autonomic Computing, Autonomic Communications and Control Loop
The self-managing future Internet is based on autonomic communications technology.  The technology is derived from control theory—the main objective of which is to manage and optimize devices and components dynamically at run time. Control theory can describe a closed system with clear architecture, but it is less able to describe open systems with uncertain information, and systems that are discrete and vary with time. Therefore it cannot be used unmodified.

    Autonomic technology is based on and expands upon control theory. It can integrate and optimize the use of various resources quickly and dynamically in an open environment. An autonomic system comprises one or more autonomic units. Each unit provides some functions and interacts with other units in a dynamic environment. An autonomic unit consists of an autonomic managing entity, and one or more managed entities. The managing entity controls the configurations, input, and output of the managed entities, which implement all system functions.

    An autonomic system can form a feedback control loop[6], as shown in Figure 1. It collects information from network measurement, environmental sensoring, users and applications’ contexts. The information is then analysed using uncertain inference, game theory, or economic modeling, and decisions are made accordingly. Finally, these decisions are acted on by the managed entities.

    Considering the increased complexity that communications and distributed systems now have, it becomes evident that the research objective of autonomic communications should be to enable a network (and its devices and services) to operate unattended with self-configuration, self-detection, self-adjustment and self-healing attributes.

    Autonomic communication technology allows the network to adjust its behavior dynamically according to user requirements. This improves network performance and resource utilization, and greatly reduces the cost of network maintenance and operation.

    Research into autonomic communications has been undertaken in areas such as Foundation, Observation, Comparison, Action, and Learning Environment (FOCALE)[7-8], Autonomic Network Architecture (ANA), and Complexity Oblivious Network Management (CONMan)[9]. However, none of these provide a general, overall autonomic network architecture. The EFIPSANS project is based precisely on autonomic communications theory. It draws on the existing research in this field

to propose a new architecture and implementation mechanism.

2 Self-Managing Network Architecture and Mechanism
The EFIPSANS project explains the future self-managing network as follows[10]:

    The basic functions of network management (such as configuration management, performance management, fault management, security management, and billing management), and basic network functions (such as routing, transfer, monitoring, and administration), can all participate in the control loop by interacting with each other automatically. This allows the network to operate and maintain itself without external intervention, and thus become a self-managing network[11].

    This definition is based on the following two assumptions[12]:

    (1) Some of the network function planes require re-construction and even integration.

    (2) In terms of network function descriptions on the nodes and devices at different levels, new concepts, function entities, and relevant framework design principles are required for implementing a self-managing network.

    Therefore, the EFIPSANS project proposes General Autonomic Network Architecture (GANA) and uses it as the basis for designing and achieving the self-managing future Internet.

2.1 General Autonomic Network Architecture
Figure 2 illustrates GANA architecture. GANA is generally designed in cubic architecture that comprises different function planes, each of which contains different levels of functional entities. In terms of functions, GANA uses the concept of 4 functional planes in 4D[13] architecture, which divides the network into decision, dissemination, discovery, and data planes. However, GANA provides specific definitions and descriptions of the functions and architecture of each plane. The architecture also defines the relationships between the planes according to the characteristics and requirements of the autonomic network.

    The decision plane controls all decisions related to network behavior including admission control, load balance, network configuration, routing, Quality of Service (QoS) and security. It also makes decisions according to network topology, flow, time, user context, network objective/policy, nodes within a specific network domain, device capability, and resource restriction changes. Thus, it controls the behavior of all the managed entities.

    The decision plane comprises different Decision Elements (DE), and together these are key components for providing autonomic properties of a self-managing network.

    The discovery plane consists of protocols or mechanisms responsible for discovering entities that make up the network or service. It is also responsible for creating logical identities to represent those entities. The discovery plane defines the scope and persistence of the identities, and carries out management accordingly. For example, it can discover how many interfaces and how many Forwarding Information Bases (FIB) a node has. It can also discover neighborhoods.

 
    The discovery plane is responsible for discovering node capabilities, networks, and services. The core of this plane consists of self-describing, self-advertizing protocols or mechanisms. In a self-managing network, these protocols or mechanisms are automatically configured by the DEs on the decision plane, and are called Managed Elements (ME).

    The data plane consists of protocols and mechanisms that handle individual packets such as IP forwarding and Layer-2 switching. This handling depends on the state that is output by the decision plane; that is, information such as FIB settings, packet filtering policy, link scheduling weight, queue management parameters, and the mapping between tunnels and network addresses. Similar to the discovery plane, the data plane also comprises multiple managed entities.

    The dissemination plane provides a reliable and efficient communication method for exchanging control information and non-user data within a node and between different node entities (such as DE and ME). There are two types of information exchange: passive acquisition (through Push) and active query (through Pull). The dissemination plane transmits signalling information, monitoring data (including status information changes), and other control information transmitted between DEs (such as fault, error, failure, and alarm information). ICMPv6, MLD, DHCPv6, SNMP, IPFIX, NetFlow and IPC all belong to this plane. The dissemination plane is also made up of multiple MEs.

    In GANA architecture, the four planes are only partially independent of each other. Some elements of the discovery plane, for example, may use data plane service and also functions provided by the dissemination plane. However, the dissemination plane may not necessarily use the data plane service. These can be independent of each other. DEs of the decision plane communicate with other through service provided by the dissemination plane.

    It should also be noted that a control loop, which is necessary for forming an autonomic system, is established between DEs and MEs on different planes. These DEs and MEs can be located on the same or different network nodes.

2.2 Decision Elements and Hierarchical Control Loop
DEs are key to GANA achieving autonomic functions and network self-management. Network functions are complex, so it is generally necessary for a node to use multiple DEs for different decision processes. In this way it can control and manage different network entities.

    As shown in Figure 3, DEs are organized hierarchically in GANA architecture. All DEs are divided according to protocol level, function level, node level, and network level. The protocol level is the lowest, involving specific protocols such as Open Shortest Path First (OSPF) and Transmission Control Protocol (TCP). Protocol-level DEs allow relevant protocols to have autonomic characteristics.

    Function-level DEs reside on the second layer. These are DEs for designing and realizing network functions. Function-level DEs abstract some specific network functions (such as routing and portability management) and the related algorithms. Node-level DEs are on the third layer. These DEs contain all the information of a network node and can affect all DEs of the node directly or indirectly. For example, the main DE of a node adjusts the node behaviors in the network by managing different DEs.

    Network-level DEs are on the highest layer. These DEs contain some of the information of other nodes in the network. They can use this information to control and affect the main DE of the local node. In addition, network-level DEs are responsible for cooperating with network-level DEs of other nodes and exchanging information with them. Thus, self-management of the entire network is realized.

    DEs in GANA architecture may have the following relationships:

• Hierarchical: a relationship between DEs on different layers within the same node;
• Peer: a relationship between DEs on the same layer between different nodes;
• Sibling: a relationship between different DEs on the same layer within the same node.

2.3 Engineering Design and Standardization Considerations
GANA is actually a reference model for the self management of nodes/devices and network architecture. It uses standardized and structural function entities, as well as specifications to ensure interoperability.

 
    In GANA architecture, the behavior triggered by a DE after it collects and analyzes information is called autonomic behavior. Autonomic behaviors, such as self-configuration, self-description, self-advertisement, self-healing and self-optimization, are those used to manage or reconfigure relevant MEs. They are also related to specific DEs; they may be bound to the information-providing parts of the control loop (where the DEs are located) or bound to each ME controlled by the DEs. Therefore, the autonomic behavior descriptions/specifications are formal, and describe the GANA architecture and DE functions.

    In addition, EFIPSANS researchers are designing models and tools for engineering DEs and the related control loops. These include meta-model, information model, system model, data model, policy model, configuration file, knowledge base, and a tool chain.

    ETSI has now established an industry specification group called Autonomic Network Engineering for the Self-Managing Future Internet (AFI-ISG)[14] for the standardization of the self-managing Internet. AFI-ISG is researching autonomic network engineering, and in particular, one of its sub-project groups is focused on GANA standardization. Another sub-project group is focused on applying the meta-model to GANA. Its objective is to determine a GANA model and to engineer control loop through formalized descriptions and designs.

3 Conclusions
Self-management will be a major feature of the future Internet. It will implement autonomic functions, including device and network self-discovery, self-configuration, resource self-provision and virtualization (without human intervention), service composition, application self-awareness, and self-monitoring. Basic network functions such as routing, forwarding, mobility management, and QoS management will also become autonomic.

 
    This article introduces GANA,developed by the EFIPSANS Project as a means of achieving self-management of the future Internet. In GANA, all autonomic functions of the network are implementd by DEs and their control loops on different planes and different layers. The next stage is to standardize and engineer GANA, which will require the description of autonomic behaviors. Research into the stability, complexity, and expandability of self-managing networks based on GANA is well underway.

References
[1] GENI Net Global Environment for Network Innovations [EB/OL]. [2009-09-30]. http://www.geni.net/.
[2] NSF NeTS FIND Initiative [EB/OL]. [2009-07-25]. http://www.nets-find.net/.
[3] European IST FP6 ANA (Autonomic Network Architecture) Project [EB/OL]. [2009-08-20]. http://www.ana-project.org/.
[4] GREENBERG A,?HJALMTYSSON G, MALTZ D A, et.al. A Clean-slate 4D Approach to Network Control and Management [J]. Computer Communication Review, 2005, 35(5): 41-54.
[5] EC funded - FP7 - EFIPSANS project [EB/OL]. [2009-09-30]. http://efipsans.org/.
[6] DOBSON S, DENAZIS S, FERNANDEZ A, et.al. A Survey of Autonomic Communications [J]. ACM Transactions on Autonomous and Adaptive Systems, 2006, 1(2): 223-259.
[7] BELL J. IBM article: Understanding the Autonomic Manager Concept [EB/OL]. [2009-09-10]. http://www-128.ibm.com/developerworks/library/ac-amconcept/.
[8] JENNINGS B, van?der MEER S,?BALASUBRANANIAN S, et al. Towards Autonomic Management of Communications Networks [J]. IEEE Communications Magazine, 2007, 45(10): 112-121.
[9] STRASSNER J C,?AGOULMINE N, LEHTIHET E. FOCALE: A Novel Autonomic Networking Architecture [C]//Proceedings of the 1st IEEE Latin American Autonomic Computing Symposium (LAACS’06), Jul 18-19,2006, Campo Grande, Brazil. Piscataway, NJ, USA: IEEE, 2006.
[10] BALLANI H, FRANCIS P. CONMan: A Step Towards Network Manageability [C]//Proceedings of Conference on Applications, Technologies, Architectures, and Protocols for Computer Communications (SIGCOMM’07), Aug 27-31, 2007, Kyoto, Japan. New York, NY, USA: ACM, 2007: 205-216.
[11] CHAPARADZA R, PAPAVASSILIOU S, ARISTOMENOPOULOS G, et al. First draft of Autonomic Behavior Specifications (ABs) for Selected Diverse Networking Environments [R]. EFIPSANS Project Deliverable D1.1v1. 2008.
[12] ITU-T Rec. M. 3400. TMN Management Functions [S]. 2006.
[13] CHAPARADZA R. The Self-managing Future Internet Powered by Current IPv6 and Extensions to IPv6 Towards IPv6++ and EFIPSANS Initiated IETF drafts [C]//Proceedings of Global Mobile Internet and IPv6: Next Generation Internet Summit 2009, Apr 15-16, 2009, Beijing, China. 2009.
[14] AFI ISG: Autonomic Network Engineering for the Self-managing Future Internet (AFI) [EB/OL]. [2009-05-30]. http://portal.etsi.org/afi/.

    Different relationships directly affect the communication mechanisms between DEs. In the GANA four-plane architecture, in order to achieve all autonomic functions, DEs on the Decision Plane and MEs on each plane form a control loop according to the principle illustrated in Figure 1. As shown in Figure 4, DEs and MEs of the nodes in GANA architecture are organized hierarchically to form a Hierarchical Control Loop (HCL).

[Abstract] This article introduces the architecture and mechanisms of the self-managing future Internet currently being developed by the European Union Seventh Framework Programme (FP7) project EFIPSANS. In this architecture, network functions are divided into four planes: decision, dissemination, discovery, and data. Decision Elements (DEs), managed entities, and other information collecting entities at different planes comprise control loops of four layers: the protocol layer, the function layer, the node layer, and the network layer. DEs and their related control loops contribute to network self-management and self-maintenance. This greatly reduces the need for intervention in the network, thus reducing costs and improving user experience. The European Telecommunications Standards Institute (ETSI) has now established a new Industry Specification Group to standardize the self-managing Internet.