3.2.1 Background
People working in NM have quickly identified deficiencies in classical NM architectures and protocols. Management data is distributed among agents with very limited process-ing capabilities, while processprocess-ing these data to extract management information is cen-tralized in managers. The latter consequently suffer from a heavy processing load. Fur-thermore, management operations must be decomposed into very primitive functions mostly limited to SET- and GET-like primitives and then communicated to the agents.
This is referred to as micro-management [YGY91]. Micro-management leads to heavy management traffic in the network.
To overcome this undesired situation, and in order to define a more flexible and scal-able architecture with real-time management capabilities, it is tacitly admitted that man-agement functions and operations should be dynamically carried out close to where the
managed objects are. Management by delegation (MbD) is designed following this prin-ciple.
One main concept on which MbD relies is that of anelastic server [Gol93]. An elastic server is an enhanced server whose functionality can be extended and reduced dynam-ically during run-time. A delegation protocol allows a client to pass new functions to an elastic server and then asks the server to execute these functions by instantiating them.
The protocol allows to control each execution instance, e.g. to stop and resume its execu-tion dynamically. New funcexecu-tionality is prescribed by a script with noa priori restriction on the language in which it is written.
As can be noticed, the MbD paradigm has been developed without initial consider-ation of software agent technologies. Nevertheless, MbD is relevant to our work at least for the following reasons:
1. Many concepts used in MbD can also be used when adopting an agent-based ap-proach. Examples of such concepts are the delegation-based cooperation and the dynamic behavior of managers.
2. A lot of work that applies MbD concepts has been or is being conducted under the banner of “intelligent software agents”.
3. We believe that this second reason was actually a driving force to consider agent technology in Network Management. Moreover, the concept of moving a script from a delegator to a delegatee is very similar to using mobile code and thereafter, mobile agent technology.
We first present works based on the MbD paradigm and how they applied the con-cepts of extensibility and delegation. Afterwards, we provide a synthesis and a discus-sion.
3.2.2 Management by Delegation
Applied to network management, elastic servers can be interpolated between managers and agents (in a similar way to middle-level managers in organizations) with similar roles to proxy agents [5]. These elastic severs are then called MbD agents [YGY91], MAD agents (Manager Agent Delegation) [YGY91], managing agents [TK97], delegation and flexible agents [Mou96] or elastic agents [GY95].
Adelegation protocolmust be defined. The manager uses the delegation protocol to upload management scripts to the flexible agents. Using the same protocol, it asks per-mission to instantiate some delegated script and to run it. Therefore, the management
operation prescribed in this script can execute “autonomously” in the same environment as the managed objects. The manager is able to control its execution via dedicated prim-itives in the delegation protocol.
A different approach than using a new delegation protocol is to make use of the ex-isting management protocols (SNMP and CMIP) in order to make classical agents evolve towards elastic agents. In this case, the process of delegating management scripts and executing them remotely is completely handled using the management protocol that is used to manipulate a special-purpose MIB designed to enable agents to receive and ex-ecute scripts. This is the approach used by the normalizing entities, i.e. the IETF and the ISO. IETF standards through the Distributed Management working group (DISMAN) pro-duced the Script MIB [LS99], while ISO standards propro-duced the Command Sequencer Systems Management Function [ISO98]. An overview of the two approaches can be found in [Sch97].
Earlier, Goldszmidt [Gol93] called the management scripts –regardless of whether they are transferred using a delegation protocol or using a network management protocol–delegated agents. We avoid using this term, which may cause confusion with the term “delegation agent”. Under the banner of intelligent agents, some research was conducted to explore the MbD paradigm for NM purposes. [KSSZ97] suggests a spread-sheet scripting environment for SNMP. The spreadspread-sheet scripting language allows a man-ager to prescribe computations that can be carried out by the agent. Each cell in the spreadsheet is defined by an expression that computes a value from other given data such as MIB variables or other existing cell values. Expressions can be inserted, updated and deleted according to the manager needs.
A further improvement is depicted in [SKN97]. Each agent contains functional ob-jects with network management functions allowing access to the management obob-jects and communication with other agents. Therefore, a manager needs only to delegate script skeletons to invoke these management functions. When running an instance, the management functions’ invocation is bound to those implemented in the functional ob-jects. The same management function could be implemented differently on different agents, e.g. according to the network device specifics.
Other works may use different scripting environments such as the SQL-like approach presented in [ZLH96], event-driven scripts presented in [Koo95] in its divide and con-quer approach and the Tcl-based scripting language in [GA97].
Alternatively, Trommer and Konopka [TK97] choose to add a rule processing unit to agents. Rules can be transmitted from managers to agents, and can endow the latter with intelligent behavior. Rules can be dynamically loaded, updated and removed, making the agents flexible yet capable of having simple and dynamic reasoning capabilities.
Finally, an interesting implementation of the MbD paradigm is that achieved by Mountzia [Mou96]. In her work, she introduces the concept of aflexible agenthaving the capability to receive scripts that define management functions and tasks. A manager or another flexible agent can therefore delegate management tasks to another flexible agent using a dedicated communication language based on KQML. In [Mou98], two delegation schemes are proposed. Inpush-based delegation, the delegatorpushesthe required func-tionality into the flexible agent; whereas inpull-based delegation, it is the flexible agent that pulls the functionality from a specified location. Flexible agents are organized hi-erarchically and they operate under the direction of a manager. A methodology is also proposed in [Mou98] and includes a task-to-subtask decomposition phase, a subtask description phase and a realization phase. This methodology and the proposed flexible agent concepts are applied in the field of fault management of distributed services. This application is described in [MR99].
3.2.3 Synthesis and Discussion of Management by Delegation
Management by Delegation is certainly a very interesting paradigm. All the approaches presented above provide many advantages even when low-level delegation such as spreadsheet-based delegation is used. One of the most appealing advantage is enhanced flexibility. The network administrator is no longer tightened to the management func-tions implemented in the NMS he uses. At runtime, the administrator can delegate new management functions and tasks to middle-level MbD agents and then invoke these functions on demand. Another advantage is the ability to bring processing capabilities close to where management data are generated, thus reducing processing overload on the management station. Finally, MbD allows to avoid the heavy management traffic due to micro-management that occurs when using a low-level management protocol such as SNMP.
The works presented above had the Management by Delegation paradigm as their starting point. For most of them the appellation “Intelligent Agent”, which actually means “MbD agent”, is mostly adopted to distinguish such agents from the normal man-agement agents, i.e. SNMP and CMIP agents. Except for the work of Mountzia and re-cently in [GY98], the other works, although using the term ‘intelligent agent’, did not show any background related to software agent technology. Later in this chapter (Sec-tion 3.6), we will have another look at the concept of delega(Sec-tion, but at a higher-level and from an agent-oriented point of view. We will show how the agent technology provides higher-level delegation than, at the end, the simple exchange of management scripts.
Management by Delegation shares the idea of bringing the code down to where
needed with Mobile Agent approaches. On the one hand, MbD can be seen as a restric-tion of MAs to the remote evaluarestric-tion paradigm in which code is uploaded to be executed close to data. On the other hand, MAs can be used, as noted in [Mou98], to transfer man-agement tasks and functionality between delegators and delegatees. In our opinion, the main conceptual difference between MbD and MAs is that in MbD, the focus is on the entities that perform delegation, i.e. the delegation agents. In contrast, the MA paradigm focuses on the mobile entities themselves and does not pay much attention to the en-tities that create and dispatch the MAs. The main difference is therefore the standpoint from which network management is considered.
The NM research community is leaving Management by Delegation in favor of the much more powerful and flexible paradigm of Mobile Agents. The following section dis-cusses the aspects of applying MA technologies in the context of NM.