Scribd
Upload
281 views5 pages

Resource Allocation in Computational Grids Using Cooperative Self-Organizing Agents

My submitted PhD research proposal that was offered the NPGS scholarship from NTU in 2009. I rejected the offer and abandoned the research. If anyone finds this interesting, go ahead and do something about it.

Uploaded by

edwinbs
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
281 views5 pages

Resource Allocation in Computational Grids Using Cooperative Self-Organizing Agents

My submitted PhD research proposal that was offered the NPGS scholarship from NTU in 2009. I rejected the offer and abandoned the research. If anyone finds this interesting, go ahead and do something about it.

Uploaded by

edwinbs
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Resource Allocation in

Computational Grids using


Cooperative Self-Organizing Agents

A Research Proposal

by

EDWIN BOAZ SOENARYO


Application Number: R0900974

submitted in partial fulfillment of the requirements


for admission to the Doctor of Philosophy graduate program
in
School of Computer Engineering
Nanyang Technological University

January 28, 2009


1. Background

The Grid infrastructure (Foster and Kesselman 1998) has allowed distributed
computational resources to be consolidated and used in solving computationally
intensive problems. In such infrastructures, allocation of finite resources across
different tasks in hand becomes a fundamental problem. This allocation is
typically handled by the service layer, which operates as a middleware between
the Grid resources and applications (Krauter, Buyya et al. 2002).

While conventional approaches of Grid resource management, such as Globus


(Foster and Kesselman 1997) and Condor (Raman, Livny et al. 1998) rely on a
centralized scheduler, recent developments have observed some decentralized
models that are based on autonomous agents, perhaps most notably ARMS (Cao
2002), the ant colony swarm algorithm (Cao 2004), and the reinforcement
learning approach (Tham and Poduval 2006). This emerging approach eliminates
the single point of failure and hence allows for higher overall reliability of the
Grid.

Despite the better reliability, under a scarcity of resources, agents may turn into
competitors for resources, as they are only evaluated individually rather than as a
group (Vyas 2004). Without proper control, such behavior is likely to render the
resource allocation inefficient and the Grid unstable. One way to address this
problem is to utilize a Grid marketplace (Lee, Choo et al. 2006), where users are
required to bid for the available computational resources. However, this approach
only prevents the resource competition situation by limiting the number of tasks,
without empowering agents to efficiently handle heavy load situations. It is
reasonable to expect self-organizing agents to be able to organize themselves and
work cooperatively under stress conditions, hence making the resource allocation
system truly robust.

2. Objective

The objective of this project is to propose ways that Grid resource allocation
agents can work cooperatively under any load condition. The proposed resource
allocation model in this research will follow the decentralized approach in order to
allow for better system reliability, while overcoming the resource competition
caveat of such approaches. The research shall pay close attention to the
implementation feasibility of the resource allocation model, as well as its
suitability for both academic and commercial Grids, which consequently requires
it to be able to incorporate dynamic Grid cost models.
Results and findings from this proposed research can remarkably improve the
efficiency and robustness of resource allocation in computational Grids.
3. Scope

In this proposed research, three main aspects of decentralized resource allocation


need to be considered:

(i) Decentralized scheduling algorithm: considers load balancing over resources


as well as how finite resource should be allocated to the outstanding tasks.
The approach may include the use of machine learning techniques as well as
probabilistic algorithms.

(ii) Cooperative negotiation protocol: explores how agents can perform group
negotiation to come out with a resource allocation plan that maximizes the
interest of the whole agent community.

(iii) Agent structure: investigates the self-organizing structure to be used in the


agent-based resource allocation model. Communication costs, especially in
the negotiation process, as well as reliability, will be considered.

This research works over the assumption that agents are developed by a single
Grid management authority, and are therefore trustworthy. Thus, it does not
consider trust and reputation in the multi-agent system.

4. Methodology

The proposed research will begin with a comprehensive literature review that
conducts a state of the art survey of the works in Grid resource allocation, studies
the notable agent negotiation methods, and explores the possible self-organizing
structures to be used in the resource allocation agent community.

This research will then progress by proposing a decentralized agent-based Grid


scheduling algorithm that moves away from the current approach of having each
agent evaluated only based on its own ability to fulfill the Service Level
Agreements (SLAs) of its tasks, but rather, agents will be evaluated as a team,
based on the whole system’s performance. With this shift of focus, there is a need
for agents to be able to negotiate with one another, and the protocols and
methods for cooperative group negotiation will be investigated next. In
conjunction with that, the agent structure to facilitate these negotiations will be
designed. Although fully centralized structures will be avoided, the proposed
structure may be a semi-centralized structure with a replaceable community
leader, or a dynamic hierarchical structure similar to that of ARMS (Cao 2002).

The proposed resource allocation model will be implemented as a prototype.


Resource allocation agents may be implemented on the JADE (Bellifemine, Caire
et al. 2003) framework, and they may utilize performance prediction toolkits such
as the WARPP toolkit (Hammond, Mudalige et al. 2009), depending on the
algorithm used. The prototype will then be empirically evaluated using a Grid
simulation software, such as GridSim (Buyya and Murshed 2002).

5. Statement of Work

The work required to fulfill the objectives of this research is expected to be


completed in three years. The breakdown of the required work can be estimated
as follows:

State-of-the art literature review 6 months


Scheduling algorithm design 6 months
Negotiation protocol design 4 months
Self-organizing agent structure design 4 months
Incorporation of cost model 3 months
Prototype implementation 4 months
Empirical evaluation and revision 3 months
Thesis preparation and writing 6 months

Total 36 months

6. Summary

Resource allocation in Grid computing directly affects the efficiency and


robustness of the infrastructure. The emergence of decentralized resource
allocation system based on self-organizing agents is expected to provide a higher
reliability by avoiding the presence of a single point of failure in the
infrastructure. Current models, however, evaluates each agents’ performance
individually without addressing the issue of potential competition for resource
among agents. This proposed research moves away from this paradigm and views
the agents as a team of resource allocator, which must work together to maximize
everyone’s benefit. The works required to come out with such a model include
designing a decentralized scheduling algorithm, proposing a cooperative group
negotiation protocol, as well as devising a structure for the resource allocation
agents. These tasks are expected to be finished within the recommended
timeframe for a Ph.D. program. Successful completion of this project will greatly
improve the efficiency and robustness of resource allocation in computational
Grids.
References
Bellifemine, F., G. Caire, et al. (2003). "JADE: A White Paper." Retrieved 23 January,
2009, from http://jade.tilab.com/papers/2003/WhitePaperJADEEXP.pdf.

Buyya, R. and M. Murshed (2002). "GridSim: a toolkit for the modeling and simulation of
distributed resource management and scheduling for Grid computing." Concurrency
and Computation: Practice and Experience 14: 1175-1220.

Cao, J. (2002). "ARMS: An agent-based resource management system for grid computing."
Scientific Programming 10(2): 135-148.

Cao, J. (2004). Self-organizing agents for grid load balancing. Fifth IEEE/ACM International
Workshop on Grid Computing (GRID 2004). Pittsburgh.

Foster, I. and C. Kesselman (1997). "Globus: A metacomputing infrastructure toolkit."


International Journal of Supercomputer Applications 11: 115-128.

Foster, I. and C. Kesselman (1998). The Grid: Blueprint for a New Computing Infrastructure,
Morgan Kaufmann.

Hammond, S. D., G. R. Mudalige, et al. (2009). WARPP: A Toolkit for Simulating High
Performance Parallel Scientific Codes. Simutools 2009, Rome, Italy, ACM SIGSIM,
SCS.

Krauter, K., R. Buyya, et al. (2002). "A taxonomy and survey of grid resource management
systems for distributed computing." Software: Practice and Experience 32: 135-164.

Lee, H. Y., T. T. Choo, et al. (2006). A Grid Market Framework. 3rd International Workshop
on Grid Economics and Business Models, Singapore, World Scientific.

Raman, R., M. Livny, et al. (1998). Matchmaking: distributed resource management for high
throughput computing. 7th IEEE International Symposium on High Performance
Distributed Computing, Chicago, Illinois.

Tham, C. K. and G. Poduval (2006). Adaptive self-organizing resource management for the
grid. 3rd International Workshop on Grid Economics and Business Models, Singapore,
World Scientific.

Vyas, T. (2004). "Grid Resource Allocation and Control using Computational Economies."
Grid Computing - Making the Global Infrastructure a Reality Retrieved 21 January,
2009, from http://www.cse.buffalo.edu/faculty/miller/Courses/Grid-Seminar/grid-
resource-allocation.pdf.

You might also like