Analyzing the tradeoffs between breakup and cloning in the context of organizational self-design By Sachin Kamboj.

Slides:



Advertisements
Similar presentations
Part IV: Memory Management
Advertisements

MINJAE HWANG THAWAN KOOBURAT CS758 CLASS PROJECT FALL 2009 Extending Task-based Programming Model beyond Shared-memory Systems.
Hadi Goudarzi and Massoud Pedram
Resource Management §A resource can be a logical, such as a shared file, or physical, such as a CPU (a node of the distributed system). One of the functions.
Using Parallel Genetic Algorithm in a Predictive Job Scheduling
Imbalanced data David Kauchak CS 451 – Fall 2013.
Meta-Level Control in Multi-Agent Systems Anita Raja and Victor Lesser Department of Computer Science University of Massachusetts Amherst, MA
XI- ADVANTAGES SCHEDULING PROCESS
Planning under Uncertainty
Resource Management of Grid Computing
Gu & Maher University of Sydney, October 2004 DECO2005 Monitoring Team Process.
Effective Coordination of Multiple Intelligent Agents for Command and Control The Robotics Institute Carnegie Mellon University PI: Katia Sycara
Decentralised Structural Reorganisation in Agent Organisations Ramachandra Kota.
Knowledge Acquisitioning. Definition The transfer and transformation of potential problem solving expertise from some knowledge source to a program.
Quality of Service in IN-home digital networks Alina Albu 23 October 2003.
Improving Robustness in Distributed Systems Jeremy Russell Software Engineering Honours Project.
Distributed Rational Decision Making Sections By Tibor Moldovan.
1 Introduction to Load Balancing: l Definition of Distributed systems. Collection of independent loosely coupled computing resources. l Load Balancing.
Architectural Design Principles. Outline  Architectural level of design The design of the system in terms of components and connectors and their arrangements.
Strategies for Implementing Dynamic Load Sharing.
“Multi-Agent Systems for Distributed Data Fusion in Peer-to-Peer Environment” Smirnova Vira ”Cheese Factory”/
Lecture Nine Database Planning, Design, and Administration
A Tool for Describing and Evaluating Hierarchical Real-Time Bus Scheduling Policies Author: Trevor Meyerowitz, Claudio Pinello, Alberto DAC2003, June 24,2003.
Database System Development Lifecycle Transparencies
WORKFLOWS IN CLOUD COMPUTING. CLOUD COMPUTING  Delivering applications or services in on-demand environment  Hundreds of thousands of users / applications.
Database Planning, Design, and Administration Transparencies
Database System Development Lifecycle © Pearson Education Limited 1995, 2005.
Engineering Law-Governed Approaches How to reuse, extend and compose interaction specifications Gustavo Carvalho, Carlos Lucena
Load Balancing in Distributed Computing Systems Using Fuzzy Expert Systems Author Dept. Comput. Eng., Alexandria Inst. of Technol. Content Type Conferences.
European Network of Excellence in AI Planning Intelligent Planning & Scheduling An Innovative Software Technology Susanne Biundo.
Use of the Concept of Transparency in the Design of Hierarchically Structured Systems Paper by: D.L. Parnas and D.P. Siewiorek Presentation by: Josh Wetzel.
An Integration Framework for Sensor Networks and Data Stream Management Systems.
SAMANVITHA RAMAYANAM 18 TH FEBRUARY 2010 CPE 691 LAYERED APPLICATION.
ANTs PI Meeting, Nov. 29, 2000W. Zhang, Washington University1 Flexible Methods for Multi-agent distributed resource Allocation by Exploiting Phase Transitions.
DLS on Star (Single-level tree) Networks Background: A simple network model for DLS is the star network with a master-worker platform. It consists of a.
Major objective of this course is: Design and analysis of modern algorithms Different variants Accuracy Efficiency Comparing efficiencies Motivation thinking.
Chapter 8 Object Design Reuse and Patterns. Object Design Object design is the process of adding details to the requirements analysis and making implementation.
By Rashid Khan Lesson 6-Building a Directory Service.
Chapter 4 Decision Support System & Artificial Intelligence.
CS 484 Designing Parallel Algorithms Designing a parallel algorithm is not easy. There is no recipe or magical ingredient Except creativity We can benefit.
1 Analysing system-user cooperation in KADS H. P. de Greef and J. A. Breuker, Department of Social Science Informatics, University of Amsterdam Knowledge.
Multiagent System Katia P. Sycara 일반대학원 GE 랩 성연식.
1 Cooperative multi-agent systems In cooperative MAS agents strive to reach a common goal and increase the combined utility of their actions Limitations.
Domain decomposition in parallel computing Ashok Srinivasan Florida State University.
Review of Parnas’ Criteria for Decomposing Systems into Modules Zheng Wang, Yuan Zhang Michigan State University 04/19/2002.
Parastoo Mohagheghi 1 A Multi-dimensional Framework for Characterizing Domain Specific Languages Øystein Haugen Parastoo Mohagheghi SINTEF, UiO 21 October.
David Chiu and Gagan Agrawal Department of Computer Science and Engineering The Ohio State University 1 Supporting Workflows through Data-driven Service.
Finite State Machines (FSM) OR Finite State Automation (FSA) - are models of the behaviors of a system or a complex object, with a limited number of defined.
1 Chapter 17 2 nd Part Making Complex Decisions --- Decision-theoretic Agent Design Xin Lu 11/04/2002.
Search Control.. Planning is really really hard –Theoretically, practically But people seem ok at it What to do…. –Abstraction –Find “easy” classes of.
Control-Theoretic Approaches for Dynamic Information Assurance George Vachtsevanos Georgia Tech Working Meeting U. C. Berkeley February 5, 2003.
Dynamic Load Balancing Tree and Structured Computations.
COMP7330/7336 Advanced Parallel and Distributed Computing Task Partitioning Dr. Xiao Qin Auburn University
COMP7330/7336 Advanced Parallel and Distributed Computing Task Partitioning Dynamic Mapping Dr. Xiao Qin Auburn University
Systems Architectures System Integration & Architecture.
MGT 200 Management Theory 4 Required Reading: Chapter 1 of textbook: Management 7th Edition by Robbins & Coulter 4 Topic: Introduction to Management.
Project Management Planning and Scheduling. “Failing to plan is planning to fail” by J. Hinze, Construction Planning and Scheduling Planning: “what” is.
1 Phase 1 Problems ● Small size – 20 to 150 methods – 10 to 20 agents ● Problem Classes – One static + one dynamically arriving problem – Uncertainty in.
The Biologically Inspired Distributed File System: An Emergent Thinker Instantiation Presented by Dr. Ying Lu.
Introduction to Load Balancing:
TÆMS-based Execution Architectures
Parallel Algorithm Design
organizational structure
Gabor Madl Ph.D. Candidate, UC Irvine Advisor: Nikil Dutt
On Spatial Joins in MapReduce
Objective of This Course
Utility-Function based Resource Allocation for Adaptable Applications in Dynamic, Distributed Real-Time Systems Presenter: David Fleeman {
Cloud Computing Architecture
Market-based Dynamic Task Allocation in Mobile Surveillance Systems
Authors: Barry Smyth, Mark T. Keane, Padraig Cunningham
Presentation transcript:

Analyzing the tradeoffs between breakup and cloning in the context of organizational self-design By Sachin Kamboj

Abstraction Organizational Self-Design (OSD) – Constructing suitable organizations at runtime – Agents are responsible for their own organizational structures Spawning a new agent – Agent is overloaded Composing agents – When they are free “Breaking" up a problem – For assigning one of the sub-problems to the newly spawned agent “Cloning" the source agent – For assigning the clone agent a portion of the source’s work load

Goal Analyze the tradeoffs between cloning and breakup and generate a hybrid model that uses both cloning and breakup to generate more suitable organizations than those that could be generated when using a single approach

Previous findings Contingency theory – There is no best way to organize – All ways of organizing are not equally effective – The optimal organizational structure depends on the problem being solved and the environmental conditions

Practically OSD is particularly suited to the problem of generating virtual organizations for grid/volunteer/cloud computing environments – Volunteer computing type of distributed computing in which computer owners donate their computing resources – Cloud computing the provision of dynamically scalable and often virtualised resources as a service over the Internet

The Model Problem solving requests (tasks) arrive at the multi-agent system at indeterminate times A single agent responsible for solving the problem in its entirety Possibilities to handle overload: – “Breakup “ - it spawns off a new agent to handle part of its load divide the problem into smaller sub- problems – “Cloning” - The individual problems are solved in their entirety by the two agents

Advantages and disadvantages Breakup – Adv Only option if the task is too “big” for any single agent Use less resources than cloning Better in situations in which the agents include a learning component – Disadv Interdependent breakup would require more coordination between the agents Cloning – Adv Only option if task cannot be broken up into smaller parts Better in executing the interdependent subtasks

TÆMS Computational framework for representing and reasoning about complex task environments Tasks are represented using extended hierarchical task structures Root represents the high-level goal that the agent is trying to achieve Sub-nodes represent the subtasks and methods that make up the high-level task Leaf nodes are at the lowest level of abstraction and represent executable methods - primitive actions

TÆMS – in more detail Executable methods – May have multiple outcomes – Different probabilities – Different characteristics Quality Cost Duration Various mechanisms for specifying subtask variations and alternatives – Characteristic accumulation function Describes how many or which subgoals or sets of subgoals need to be achieved in order

TÆMS – In real world Distributed sensor networks Information gathering Hospital scheduling EMS - Engine Management System Military planning

Task Structure – T - set of tasks, the non-leaf nodes of a TÆMS task structure with quality and quality/characteristic accumulation function – τ - root of the task structure – M – set of primitive methods that can be directly executed by an agent With the probability that the outcome will have a quality/cost/duration – Q - is the set of quality/characteristic accumulation functions – E - set of (non-local) effects – R - set of resources – ρ - mapping from an executable method and resource to the quantity of that resource needed (by an agent) to schedule/execute that method – C - mapping from a resource to the cost

OSD An evaluation component – Monitoring the performance An adaptation component – Triggered by the evaluation component when the utility falls below a threshold to modify the organizational structure

Local Task Structure Obtained by rewriting the global task structure and represent the local task view of the agent vis-a-vis its role in the organization and its relationship to other agents No single agent has a global view of the complete organization

organizational nodes Two flavors – Organizational tasks Used to aggregate other organizational nodes – Organizational methods Represent either organizational knowledge or organizational actions that have some fixed semantics Types – Container-Nodes Aggregates of domain nodes and other organizational nodes – Non-Local-Nodes Represent nodes in the global task structure that the agent knows the identity of but does not know the characteristics – Clone Selectors Used to select amongst the clones of a node – NLE-Inheritors Transfer the non-local effect from a non-cloned node to a cloned node or vice versa

Organization Structure Breakup – Need to consider interrelationships (NLEs) – The two agents can negotiate a coordination – Non-local-nodes will be added to the root-node Merging – Exact inverse of breakup – Should be associative Merge order is not important Cloning – Situation Breakup might be infeasible The agent would prefer to do simple load balancing – Use Clone container to hold all the created nodes Clone selector to enable one or more clones – Careful handle of boundaries Preserve their original semantics Allow the presence of clones to be transparent to the non clone nodes

Spawning Strategies Breakup – Task structure is always broken up into smaller subtasks – Used the Balancing Execution Time (BET) heuristic to select the node to be allocated to the new agent Prefer Breakup – Same as breakup but with the exception that if Breakup is infeasible, the Cloning approach is used

Spawning Strategies … Cloning – The root of the task structure is always cloned and assigned to the newly spawned agent Prefer Cloning – Similar to the Cloning approach, with the exception that if Cloning is infeasible given the current task load, the agent will Breakup according to the BET heuristic

Spawning Strategies … Hybrid Model – A combination of cloning the highest level goal and breakup according to the BET heuristics – Utility value, which is the expected utility of breaking up according to the BET heuristics – If utility is greater than a constant called the breakup threshold, the agent chooses to breakup – Otherwise it clones the highest level node