Managing risks in an open computing environment using mean absolute deviation portfolio optimization
Introduction
An open computing environment is a new paradigm for using computing resources. Until now, in most cases, users who needed computing resources have bought and maintained their own resources. While scientific users in the academic area sought other alternatives, for example, renting computing resources from supercomputing centers or implementing shared infrastructures, in order to obtain high-performance computing power, commercial area users purchased additional proprietary IT facilities for intensive computational works [1], [2]. In this case, the amount of computing resources purchased was usually larger than the average usage, and therefore users suffered cost inefficiency. Moreover, even if there were urgent and temporary excess demands for some computing resources, those requests were bounded by the location and capacity of physical computing resources. In an open computing environment, which has been enabled by Grid computing technologies, however, various kinds of computing resources are provided from networks with different levels of performance and price. Users can buy exactly as many computing volumes as they request. This new computing structure has enabled much more flexibility in the cost, capacity and mobility of computing resources [3].
From the early stage of Grid computing, there have been active discussions about how to transform the volunteer-based sharing infrastructure into a market-based trading system for commercial use [4], [5]. Because the Open Grid Service Architecture (OGSA) has been established, the transformation seemed to be at hand. However, it has been difficult for the computing resource market to settle down in the best effort system, in which the current system resides [6], [3]. That is because the best effort system like the Internet does not support hard quality of services (QoS) for the resources’ performance. The commercialization of Grid computing or Grid services is intended to transform computing resources, such as hardware, software and data, into commodities tradable in a market. Commoditization implies the supply of a certain quantity of resource blocks to buyers. In this context, even resources of poor performance are more appropriate to use than those of uncharacterized performance [3]. However, computing resources supplied in the open computing environment hardly have constant performance because of the dynamic characteristics of this environment [7], [3]. They have variances or fluctuations in their performance, which are risks to users. Therefore, performance prediction became one of the main requirements for managing Grid environments [8]. Much research, which aimed to design a representative Grid simulation tool, also addressed this dynamic aspect of Grid environment and tried to incorporate in their modeling [9], [10], [11]. According to [2], [12], the Grid computing environment evolves toward a utility computing that is highly diversified and heterogeneous in systems as well as providers. This trend is expected to increase the intensity of dynamic characteristics in Grid environments. As sellers and buyers cannot predict the exact performance of computing resources that they trade, users cannot trust the price of the market. Users’ participation will be thwarted by those uncertainties. Therefore, managing the performance variance of open computing resources is a crucial issue to be addressed in order that the open computing market can grow.
In this paper, we investigate how to manage resource uncertainties when running a job in an open computing environment. Our model is expected to enhance the commercialization of open computing resources, which eventually help users to obtain more flexible and ubiquitous access to computing resources and to take more opportunities in businesses, lifestyle, and social activities. In the real world, portfolio diversification has long been used for controlling risks. The risk management framework proposed in this paper adopts such a real-world practice. An open computing broker uses the mean absolute deviation (MAD) portfolio optimization model to provide reliable resource provisioning without suffering from the performance fluctuation of individual resources.
One important question in using this method is how a portfolio composer gathers necessary data to estimate the variance of a volatile resource. First, a broker may utilize the computing resources’ performance data stored in a centralized directory service of the typical Grid computing architecture [13]. Though it provides service brokers with high efficiency to gather data about open computing resources, brokers need to make an extra effort to estimate the risk of each open computing resource, as the global data repository stores only the average performance data. The other disadvantage of using the directory service is that the data registered might be false or out of date. With regard to obtaining the risk information, this study suggests two alternative methods: real-time update and direct data management. To verify the effect of our MAD model on risk management in an open computing environment, we used the simulation methodology. The simulation results showed that the MAD model could successfully control the risks of open computing resources.
The remainder of this paper is structured as follows. Section 2 introduces previous works related to risk management in open computing environment and MAD portfolio optimization. Section 3 describes the modeled framework of an open computing market. A broker’s problem to provide a reasonable level of reliability with open computing resources is also designed in Section 3. Section 4 explains the simulation scenarios to test the performance of the proposed MAD model. The results of simulations are displayed in Section 5. Lastly, Section 6 concludes with the analysis of simulation results and discussions.
Section snippets
Related works
The majority of early discussions on an open computing market were pricing, accounting, billing, and payment [4]. As the construction of a reliable and secure system is regarded as the key factor for the success of an open computing market [5], particular attention has been directed to such technologies as security [14], [15], advanced reservation [16], and sabotage tolerance [17], which improve the reliability of the open computing system for commercial use. However, within the current best
Model
This section lays out our risk management model for an open computing market using the reformulated MAD portfolio optimization technique.
Simulation
We developed eight models to investigate the effectiveness of the MAD portfolio optimization model as a risk management tool for an open computing environment in various conditions. Table 1 describes the characteristics of eight models. They differ in respect of whether they adopt the risk management model and the source of the resource information. Model 1 is a basic model in which unknown resources are randomly assigned to users’ job requests. A broker does not consider adopting any risk
Results
First, we display the ways in which the two data collection strategies proposed in Section 3.4 work differently in the implemented system. For the indirect data collection strategy, which was applied to Models 2–7, all the open computing resources are known resources for the entire simulation periods. As a service broker utilizes the existing architecture, data for all the open computing resources in the VO are ready from the very beginning, to be utilized for composing portfolios. On the other
Discussion
Throughout the simulations, open computing environments with the MAD portfolio optimization model could establish stable and prosperous markets, whereas the “do nothing” alternative, Model 1, ended up with market failure in risky environments. That implies that diversified resource portfolios created by the MAD method are capable of managing risks in an open computing environment. The important finding of our research was that, in order for service brokers to provide reliable services to users,
Acknowledgements
This research was supported by the KCC (Korea Communications Commission), Korea, under the CPRC (Communications Policy Research Center) support program supervised by the IITA (Institute for Information Technology Advancement) (IITA-2009-(C1091-0901-0003)).
Junseok Hwang is an Associate Professor in Technology Management, Economics and Policy Program of Seoul National University in Korea. Prior to this, he was an Assistant Professor in the School of Information Studies at Syracuse University. He received his Ph.D. in Information Science and Telecommunications from the University of Pittsburgh. His research interests include network economics, next generation internet, network convergence theory and social effect of network technologies. He may be
References (40)
- et al.
Resource trading using cognitive agents: A hybrid perspective and its simulation
Future Generation Computer Systems
(2007) - et al.
Dynamic security perimeters for inter-enterprise service integration
Future Generation Computer Systems
(2007) - et al.
Sabotage-tolerance and trust management in desktop grid computing
Future Generation Computer Systems
(2007) - et al.
A trust-based consumer decision-making model in electronic commerce: The role of trust, perceived risk, and their antecedents
Decision Support Systems
(2008) - et al.
I-TRUST: Investigating trust between users and agents in a multi-agent portfolio management system
Electronic Commerce Research and Applications
(2003) - et al.
A commodity market algorithm for pricing substitutable grid resources
Future Generation Computer Systems
(2007) Chapter 16 agent-based computational economics: A constructive approach to economic theory
A Networking Approach to Grid Computing
(2004)- et al.
Grid Computing: The Savvy Manager’s Guide
(2005) - et al.
Grid resource commercialization: Economic engineering and delivery scenarios
Grid services for distributed system integration
Computer
Experiences with predicting resource performance on-line in computational grid settings
SIGMETRICS Performance Evaluation Review
Realistic modeling and svnthesis of resources for computational grids
Synthesizing realistic computational grids
The MicroGrid: A scientific tool for modeling computational grids
Scientific Programming
Adopting the grid for business purposes: The main objectives and the associated economic issues
Grid Economics and Business Models
The anatomy of the grid: Enabling scalable virtual organizations
International Journal of High Performance Computing Applications
Trust and Risk in Internet Commerce
Cited by (0)
Junseok Hwang is an Associate Professor in Technology Management, Economics and Policy Program of Seoul National University in Korea. Prior to this, he was an Assistant Professor in the School of Information Studies at Syracuse University. He received his Ph.D. in Information Science and Telecommunications from the University of Pittsburgh. His research interests include network economics, next generation internet, network convergence theory and social effect of network technologies. He may be reached at [email protected].
Hak-Jin Kim is an assistant professor in the school of business of Yonsei University in Korea. He received his Ph.D. in Operations Research from Carnegie Mellon University. His research interests include hybrid approach of optimization and constraint programming, product line design in production, semantic web using OR techniques and option bound problems in investment. He may be reached at [email protected].
Jihyoun Park is a Ph.D. candidate of Technology Management, Economics and Policy Program of Seoul National University in Korea. She received her Bachelor’s degree in Computer Science and Engineering from Seoul National University. She has been involved in several projects about developing Grid business models in Korea and EC. Her current research interests are Grid economics and business models, risk management and trust-building mechanisms in open and autonomous computing environments. She may be reached at [email protected].