Engineering of IT managment automation along task analysis, loops, function allocation, machine capabilities [Elektronische Ressource] / vorgelegt von Ralf König

Ralf König

Sujets

• Informatik

Informations

• Publié par: ludwig-maximilians-universitat_munchen
• Publié le: 01 janvier 2010
• Nombre de lectures: 8
• Langue: English
• Poids de l'ouvrage: 11 Mo

Extrait

Engineering of IT Management Automation along Task
Analysis, Loops, Function Allocation, Machine Capabilities
Dissertation
an der
Fakultat¨ fur¨ Mathematik, Informatik und Statistik
der
Ludwig-Maximilians-Universitat¨ Munchen¨
vorgelegt von
Ralf Konig¨
Tag der Einreichung: 12. April 2010Engineering of IT Management Automation along Task
Analysis, Loops, Function Allocation, Machine Capabilities
Dissertation
an der
Fakultat¨ fur¨ Mathematik, Informatik und Statistik
der
Ludwig-Maximilians-Universitat¨ Munchen¨
vorgelegt von
Ralf Konig¨
Tag der Einreichung: 12. April 2010
Tag des Rigorosums: 26. April 2010
1. Berichterstatter: Prof. Dr. Heinz-Gerd Hegering, Ludwig-Maximilians-Universitat¨ Munchen¨
2. Prof. Dr. Bernhard Neumair, Georg-August-Universitat¨ Gottingen¨Abstract
This thesis deals with the problem, that IT management automation projects are all tackled in a different
manner with a different general approach and different resulting system architecture.
It is a relevant problem for at least two reasons: 1) more and more IT resources with built-in or asso-
ciated IT management automation systems are built today. It is inefficient to try to solve each on their
own. And 2) doing so, reuse of knowledge between IT management automation systems, as well as
reuse of knowledge from other domains is severely limited. While this worked with simple stand-alone
remote monitoring and remote control facilities, automation of cognitive tasks will more and more
profit from existing knowledge in domains such as artificial intelligence, statistics, control theory, and
automated planning. A common structure also would ease integration and coupling of such systems,
delegating cognitive partial tasks, and switching between commonly defined levels of automation.
So far, this problem is only partly solved. Prior work on approaching systems design includes systems
engineering with its steps task analysis and function allocation. But so far systems engineering has
not yet been widely applied to IT management automation systems, in contrast to technical application
domains such as automotive or aerospace engineering.
The state of the art in IT management automation itself on the technical side includes IBM’s Auto-
nomic Computing Initiative, proposing a structure of feedback loops for autonomic systems. On the
organizational side, a certain standardization of IT service management processes has been reached
with the introduction of process management standardization work like the IT infrastructure library
(ITIL), which is a collection of best practices in IT service management, and ISO 20.000, a standard
derived from ITIL V2. So far, the mentioned developments all stand each on their own.
The idea of this thesis is now to combine this prior knowledge to create a universal approach to the
design of IT management automation systems in four steps:
Step 1: Task analysis in the IT management automation scenario, with an implicit goal to associate the
identified tasks with a subset of the ITIL reference processes (see Chapter 3).
Step 2: Reformulation of the tasks into feedback loops of a common structure for cognitive loops (see
Chapter 4).
Step 3: Function allocation of the loop steps to either performed by humans or machine components
(see Chapter 5).
Step 4: Identification of a set of relevant machine capabilities in a catalog structured along the previous
three steps (see Chapter 6).
Each of the steps 1–3 is the result of combining existing knowledge in systems engineering and existing
knowledge in IT management automation.
During this combination the author of this thesis also creates new structures, such as an automation-
oriented subset of ITIL V2 called “AutoITIL”, the “MAPDEK” (monitor, analyze, plan, decide, execute
based on knowledge) loop as basic loop and simpler ways of function allocation to humans/machines.
The catalog of machine capabilities, presented in Step 4 then adds a non-exhaustive catalog of machine
capabilities in IT management automation. It is domain-specific by design and clearly exceeds similar
more general attempts such as Fitt’s “Men are better at, machines are better at” list, 1951. Otherconcepts are taken over unaltered after a review during this combination, such as proposals by Sheridan
(2000) on the structure and description of tasks and levels of automation. All in all, this way we gain a
domain-specific method for engineered IT management automation (EIMA).
The EIMA method is intended to be used in the analysis and design of IT management automation sys-
tems. Its purpose is to first align the structure of the automation problem to EIMA and then to quickly
identify automation potential enabled by existing mathematical methods and machine capabilities.
In an application example (see Chapter 7), it is shown how an existing complex proposal for partial
automation in a wide area network scenario is analyzed, and a new proposal is derived that improves along known concepts of control systems, and at the same time simplifies the organiza-
tional structure. It does so by decomposing the overall problem into a set of smaller problems that
can be solved by domain experts in economics, graph theory, network configuration management. A
comparison shows that while the previous proposal has created a UML-based draft model for a pro-
tocol that eases information exchange, the new proposal actually includes the cognitive tasks that had
been left to system administrators in the first proposal. A variable level of automation leaves enough
influence by network operators on the automation routine, and therefore provides the basis for better
acceptance.
This work differs from related work in not focusing on single automation scenario instances (regarded
as point solutions), but instead proposing a general method applicable to all such problems and bringing
a benefit by its association with existing concepts in systems engineering and IT management. As the
method is partly rooted in systems engineering, we can build on the experience that has been gathered
there in the design of control systems like autopilots and driver assistance systems.
The main benefit of EIMA for system designers is a uniform way of describing and decomposing IT
management automation problems even of high complexity at a high level of abstraction. Only after
this high-level decomposition, system designers and domain specialists will take care of the details by
further refining the tasks and implementing system components. This top-down approach makes even
complex IT management automation systems comprehensible and concise in their description. Also
for system administrators this is a clear benefit.
To achieve this simplicity, EIMA also must leave certain aspects unconsidered. Due to its universality,
we can hardly include scenario-specific aspects: EIMA makes no advice on the “best” automation.
Instead, it leaves out cost benefit analysis or return-on-investment analysis as this
depends on external policies as well. EIMA also does not give detailed cookbook-like advice on which
machine capabilities to use for what task. In this point it must rely on the experience, background
knowledge and creativity of a user of the EIMA method, all of which is supported by EIMA, but not
guided along a single path of cause effect relationships to one particular systems design.
Potential follow-up work includes the creation of tools for EIMA itself: modeling tools, a knowledge
base of machine capabilities, the electronic representation of partial solutions such as loop patterns,
and simple visualization and simulation. In the same way as the design of electronic circuits today is
supported with circuit simulation programs such as SPICE, or design of 3D objects such as machines
or buildings with CAD applications such as AutoCAD. Once such tools were available, EIMA models
could be exchanged in terms of formal, machine-interpretable files instead of mainly verbal system
descriptions interpreted by humans.Zusammenfassung
Diese Dissertation beschaftigt¨ sich mit dem Problem, dass Projekte zur IT-Management-Automati-
sierung jedes fur¨ sich anders angegangen werden und daraus unterschiedliche Systemarchitekturen
entstehen.
Die entstehende Heterogenitat¨ ist aus mindestens zwei Grunden¨ problematisch: 1) Immer mehr IT-
Systeme mit eingebauten oder gekoppelten IT-Management-Automatisierungsfunktionen werden heute
entworfen. 2) Dabei wird nur in sehr begrenztem Maße auf vorhandenes Wissen aus anderen IT-
Management-Automatisierungsystemen oder verwandten Wissensfeldern zuruckge¨ griffen. Wahrend¨
¨der bestehende Ansatz fur¨ einfache Uberwachungssysteme und Fernsteuersysteme noch ausreichte,
wurde¨ die Automatisierung der kognitiven Aufgaben immer mehr von Methoden aus den Gebieten
Kunstliche¨ Intelligenz, Statistik, Steuer- und Regelungstechnik, Automatisiertes Planen profitieren.
Eine einheitliche Struktur von IT-Management-Automatisierungssystemen wurde¨ auch die Integration
und Kopplung solcher Systeme vereinfachen, sowie die Delegation von kognitiven Teilaufgaben und
das Umschalten zwischen einheitlich definierten Automatisierungsgraden.
Bisher ist dieses Problem nur teilweise gelost.¨ Ein umfassender Ansatz zu Systemanalyse und Sys-
tementwurf wird im Systems Engineering beschrieben mit den Teils