Middleware 06  
4th International Workshop on Middleware for Grid Computing - MGC 2006

In conjunction with ACM/IFIP/USENIX 7th International Middleware Conference 2006
Melbourne, Australia - November 27 December 1, 2006

Time Title
8:45AM Opening
 
Introduction to Middleware for Grid Computing
Bruno Schulze

 
9:15AM Session 1   Slides Text
 
David Chadwick, Linying Su, Romain Laborde

 
 
Yash Patel, John Darlington

   
 
Lei Ni, Aaron Harwood, Peter Stuckey

   
10:45AM Coffee-Break
11:15AM Session 2
 
Rüdiger Kapitza, Meik Felser, Hans Reiser, Franz Hauck

 
 
Xingchen Chu, Lonie, Peter Harris, Randy Thomas, Rajkumar Buyya

 
 
Eddy Caron, Andreea Chis, Alan Su

 
12:45AM Lunch
02:15PM Session 3
 
José de Ribamar Pinheiro Jr., Fabio Kon, Alexandre Vidal, Marcelo Finger

 
 
Jiaying Zhang, Peter Honeyman

 
03:45PM Session 4
 
Stephan Hirmer, Hartmut Kaiser, Andre Merzky, Andrei Hutanu, Gabrielle Allen

 
 
Luiz Bittencourt, Edmundo Madeira

 
 
Andrei Hutanu, Stephan Hirmer, Gabrielle Allen, Andre Merzky

 
05:15PM Session 5
 
Paulo Silva, Carlos Westphall, Marcos Assuncao, Carla Merkle Westphall

 
 
Po-Cheng Chen, Jyh-Biau Chang, Tyng-Yeu Liang, Ce-Kuen Shieh, Yi-Chang Zhuang

 
 
Francisco Brasileiro, Lauro Costa, Alisson Andrade, Walfredo Cirne, Sujoy Basu, Sujata Banerjee

 
03:15PM Postres & Coffee-Break
 
Carlos Queiroz, Marco Netto, Rajkumar Buyya

 
 
Alexandre Vidal, José de Ribamar Pinheiro Jr., Fabio Kon, Sergio Kofuji

 
 
HongSoo Kim, EunJoung Byun, ChongSun Hwang

 
 
Agustín Caminero Herráez, Carmen Carrión Espinosa, Blanca Caminero Herráez

 
 
Jianjun Hu, Heqing Guan, Hua Zhong

 
 
Asif Akram, Robert Allan

 
 
A.B.M. Russel, Asad I. Khan, David Abramson

 


Providing Secure Coordinated Access to Grid Services

David Chadwick, Linying Su, Romain Laborde, University of Kent, Great Britain.

  Abstract: Coordinating the use of distributed resources in a grid environment so that users do not consume too much is a difficult task. This paper presents one approach that we have implemented in Globus Toolkit version 4 (GT4), that uses an SQL database to hold coordination data, and a policy decision point (PDP) to make access control decisions about whether the users request for more resources can be granted or denied. When access is granted, obligations in the policy ensure that the coordination database is appropriately updated. In our initial implementation, the coordination service is imbedded into the GT4 authorisation chain as a custom PDP so that any a web service canbe provided with a security policy that provides a coordination capability. In the final section we describe how coordinated decision making could be more tightly integrated into a future version of Globus Toolkit.  
A Novel Approach To Allocating QoS-constrained Workflow-based Jobs In A Multi-Cluster Grid
Yash Patel, London e-Science Centre, Imperial College, Great Britain
John Darlington, London e-Science Centre, Imperial college, Great Britain.
  Abstract: Clusters are increasingly interconnected to form multicluster systems, which are becoming popular for scientific computation. Grid users often submit their applications in the form of workflows with certain Quality of Service (QoS) requirements imposed on the workflows. These workflows detail the composition of Grid services and the level of service required from the Grid. This paper addresses workload allocation techniques for Grid workflows. We model a resource within a cluster as a G/G/1 queue and minimise failures (QoS requirement violation) of jobs by solving a mixed-integer non-linear program (MINLP). The novel approach is evaluated through an experimental simulation and the results confirm that the proposed workload allocation strategy not only provides QoS guarantee but also performs considerably better in terms of satisfying QoS requirements of Grid workflows than reservation-based scheduling algorithms.  
Realizing the e-Science Desktop Peer using a Peer-to-Peer Distributed Virtual Machine Middleware
Lei Ni, Aaron Harwood, Peter Stuckey, University of Melbourne, Australia.
  Abstract: Emerging e-Science applications face the challenge of providing both high performance and low maintenance cost infrastructure. We propose the e-Science Desktop Peer system and a prototype middleware implementation called P2P-DVM to address this challenge, that is designed to allow scientists'' resources to be readily shared over the Internet in a Peer-to-Peer (P2P) manner. Our P2P-DVM currently supports the Message Passing Interface and Bulk Synchronous Processing model over an adaptive P2P network and we introduce a Programming Environment Abstraction Layer to allow users to quickly adapt other programming environments to our P2P approach. P2P-DVM provides decentralized coordinated checkpoint and restart functionality, message passing over P2P networks, and distributed storage for scalability. Our experimental results on the PlanetLab testbed shows support for our work.  
Decentralised diffusion-based Quota Management
Rüdiger Kapitza, FAU Erlangen, Germany.
  Abstract: Abstract. Gird infrastructures often rely on centralised services for resource management and quota enforcement. This is an efficient approach for small and midsize girds but might turn into a bottleneck and requires extensive hardware and software application in case of large-scale infrastructures. Further on additional installation and maintenance work is needed to provide a fault-tolerant solution. This paper presents a decentralised, fault-tolerant and scalable quota enforcement protocol. By utilising the concept of diffusive load balancing unused resource quotas dedicated to projects or jobs are equally balanced over all supporting nodes. This enables the enforcement of upper resource bounds while being adaptive to changing resource demands.  
KidneyGrid: A Grid Platform for Integration of Distributed Kidney Models and Resources
Xingchen Chu, University of Melbourne, Australia
Lonie, Peter Harris, The University of Melbourne, Australia
Randy Thomas, Université d'Evry, France
Rajkumar Buyya, University of Melbourne, Australia.
  Abstract: In this paper, we present a grid computing platform that provides experimental scientists and analysts with access to computational simulations and knowledge databases hosted in separate laboratories around the world involved with human and animal kidney research. No single laboratory can develop these resources in isolation and the community of users should no longer need to be dependent upon the specific programming environment in which applications have been developed. This is a major innovation in life science and exploits the power of existing and planned high-bandwidth communications networks for collaborative research and for shared access to knowledge resources. The innovation is developed within a specialist community of renal scientists but will be transferable to any other field of research requiring interaction between published literature and databases, theoretical models and simulations and the formulation of effective experimental design.  
Plug-in Scheduler Design for a Distributed Grid Environment
Eddy Caron, Andreea Chis, Frederic Desprez, ENS-Lyon / INRIA, France
Alan Su, Google, USA.
  Abstract: This paper presents the approach chosen within the DIET project to allow a resource broker to be tuned for specific application classes. Our design allows the use of generic or application dependent performance measures in a simple and seamless way. We introduce the architecture of the DIET middleware framework. Thus, we can describe the plug-in scheduler feature and the CORI collector which allows the management of different performance measures. To validate this works we present some early experiments of this new feature of our GridRPC framework. 
Trust in Large-Scale Computational Grids: An SPKI/SDSI Extension for Representing Opinion
José de Ribamar Pinheiro Jr., Universidade de São Paulo, Brazil
Fabio Kon, Alexandre Vidal, IME-USP, Brazil
Marcelo Finger, Universidade de São Paulo, Brazil.
  Abstract: SPKI/SDSI is a flexible and extensible decentralized security model that provides authentication, confidentiality, and access control. However, SPKI/SDSI certification chains are not suitable for large scale, highly dynamic environments such as computational grids. This work extends the SPKI/SDSI model by including an opinion model based on subjective logic. A simulation is performed to evaluate the effectiveness of the proposal.  
NFSv4 Replication for Grid Storage Middleware
Jiaying Zhang, University of Michigan, USA
Peter Honeyman, University of Michigan, USA.
  Abstract: Sharing data in scientific collaborations that involve many institutions around the world demands a large-scale storage system that is reliable and efficient, yet at the same time, convenient to use. This paper presents a replicated file system that supports mutable replication with strong consistency guarantees, high failure resiliency, and good scaling properties. The principal concern with such a system is usually its performance penalty. Using experiment evaluation, we show that in simulated wide area networks, the proposed replicated file system maintains a significant performance advantage over a single server system. At the same time, it provides comparable and often better performance than GridFTP, the de facto standard file transfer application that is often used to manually synchronize shared data.  
Generic Support for Bulk Operations in Grid Applications
Stephan Hirmer, Hartmut Kaiser, CCT@LSU, USA
Andre Merzky, Vrieje Universiteit Amsterdam, The Netherlands
Andrei Hutanu, Louisiana State University, USA
Gabrielle Allen, CCT@LSU, USA.
  Abstract: Within grid environments, latencies for remote operations of any kind can, as the number of operations increases, become a dominant factor for overall application performance. Amongst various approaches for latency hiding, bulk operations provide one possible solution to reduce latencies for large numbers of similar operations. The identification of bulks can, however, pose a non-trivial exercise for application developers, often requiring changes to the implemented remote API, and hence direct code modifications to the application themselves. In this paper we show how bulk operations can be integrated into existing API implementations, and identify the required properties of the API to make this approach feasible. We also show that our approach considers any type of bulk operation, and is independent of the underlying middleware support for bulks. We further describe a prototype implementation (within the SAGA C++ reference implementation effort), and present performance measurements for bulks of remote file copy operations. 
A Dynamic Approach for Scheduling Dependent Tasks on the Xavantes Grid Middleware
Luiz Bittencourt, Edmundo Madeira, UNICAMP, Brazil.
  Abstract: A scheduler must consider the heterogeneity and communication delays when scheduling dependent tasks on a grid. The task scheduling problem is NP-Complete in general, what led us to the development of a heuristic for the associated optimization problem. In this work we present a dynamic approach to schedule dependent tasks onto a grid based on the Xavantes grid middleware. The developed dynamic approach is applied to the Path Clustering Heuristic (PCH), and introduces the concept of rounds, which take turns sending tasks to execution and evaluating the performance of the resources. The experiments show that the round-based dynamic schedule can minimize the effects of performance losses while executing processes on the grid. 
Analysis of Remote Execution Models for Grid Middleware
Andrei Hutanu, Louisiana State University, USA
Stephan Hirmer, Gabrielle Allen, CCT@LSU, USA
Andre Merzky, Vrieje Universiteit Amsterdam, The Netherlands.
  Abstract: Grid computing applications, and distributed applications in general, often experience performance deterioration due to the latencies inherent in the execution of remote operations. Here we analyze three approaches for reducing latencies: an asynchronous model which executes operations in a thread to hide the remote latency of an operation, a bulk model which bundles multiple operations together in a single remote operation, and a pipelining model which executes remote operations in a pipeline-parallel mode. We analyze the performance, parameters and technical requirements of each model, and identify general properties which can help determine which model is the most suitable. Our results show that, depending on the use case scenario, any of the three models can o er the best performance, and we conclude by presenting a hybrid model that combines all three approaches, potentially providing the benefit of each.
Composition of a DIDS by Integrating Heterogeneous IDSs on Grids
Paulo Silva, Universidade Federal de Santa Catarina, Brazil
Carlos Westphall, Federal Uiversity of Santa Catarina, Brazil
Marcos Assuncao, University of Melborne, Australia
Carla Merkle Westphall, UFSC, Brazil.
  Abstract: This paper considers the composition of a DIDS (Distributed Intrusion Detection System) by integrating heterogeneous IDSs (Intrusion Detection Systems). A Grid middleware is used for this integration. In addition, an architecture for this integration is proposed and validated through simulation.  
A Multi-layer Resource Reconfiguration Framework for Grid Computing
Po-Cheng Chen, Institute of Computer and Communication Engineering, Department of Electrical Engineering, National, Republic of China
Jyh-Biau Chang, Department of Information Management, Leader University+, Republic of China
Tyng-Yeu Liang, Department of Electrical Engineering, National Kaohsiung University of Applied Science, Republic of China
Ce-Kuen Shieh, Department of Electrical Engineering, National Cheng Kung University, Republic of China
Yi-Chang Zhuang, Home Network Technology Center, Industrial Technology Research Institute/South, Republic of China.
  Abstract: Grid is a non-dedicated and dynamic computing environment. Consequently, different programs have to compete with each other for the same resources, and resource availability varies over time. That causes the performance of user programs are degraded and unpredictable. For resolving this problem, we propose a multi-layer resource reconfiguration framework for grid computing. As the named, this framework adopts different resource reconfiguration mechanisms for different workloads of resources. We have implemented this framework on a grid-enabled DSM system called Teamster-G. Our experimental result shows that our proposed framework allows Teamster-G not only to fully utilize abundant CPU cycles but also to minimize resource contention between the jobs of resource consumers and those of resource providers. As a result, the job throughput of Teamster-G is effectively increased.  
A large scale fault-tolerant grid information service
Francisco Brasileiro, Universidade Federal de Campina Grande, Brazil
Lauro Costa, UFCG - Universidade Federal de Campina Grande, Brazil
Alisson Andrade, Walfredo Cirne, Universidade Federal de Campina Grande, Brazil
Sujoy Basu, Hewlett-Packard Labs, USA
Sujata Banerjee, HP Labs, USA.
  Abstract: Large scale grid systems may provide multitudinous services, from different providers, whose quality of service will vary. Moreover, services appear (and disappear) in the grid with no central coordination. Thus, to find out the most suitable service to fulfill their needs, grid users must resort to Grid Information Services (GISs). These services allow users to submit rich queries that are normally composed of multiple attributes and range operations. The ability to efficiently execute complex searches in a scalable and reliable way is a key challenge for current GISs. Scalability issues are normally dealt with by using peer-to-peer technologies. However, the more reliable peer-to-peer approaches do not cater for rich queries in a natural way. On the other hand, approaches that can easily support these rich queries are less robust in the presence of faults. In this paper we focus on peer-to-peer GISs that efficiently support rich queries. In particular, we thoroughly analyze the impact of faults in one representant of such GISs, named NodeWiz. We propose extensions that increase NodeWiz''s resilience to faults.
POSTERS
Message Passing over Windows-based Desktop Grids
Carlos Queiroz, Marco Netto, Rajkumar Buyya, University of Melbourne, Australia
  Abstract: Message Passing is a mechanism to enable inter-process communication in parallel and distributed computing. Complex scientific and engineering applications have been developed upon such a communication model to be executed on cluster machines. Due to the huge amount of computer power being wasted in desktop machines, there is an increasing interest in using idle machines to execute these applications. However, most of the current middleware systems are aimed at executing only embarrassingly parallel applications, i.e. with no inter-process communication. Moreover, incredibly, these existing systems are based on UNIX-like when we know that the very most of the desktop machines around the world are based on Windows. In this work we present the design, implementation and evaluation performance of a Windows-based implementation of two message passing models, Message Passing Interface (MPI) and Bulk Synchronous Parallel (BSP), over the Alchemi''s Grid middleware.
Defining and Exploring a Grid System Ontology
Alexandre Vidal, IME-USP, Brazil
José de Ribamar Pinheiro Jr., Universidade de São Paulo, Brazil
Fabio Kon, IME-USP, Brazil
Sergio Kofuji, LSI - EPUSP, Brazil.
  Abstract: This work introduces a generic and extensible grid system ontology and presents an approach to explore it in different grid scenarios. We explore a simple and effective way for developing the ontology, driven by use cases analysis. We show the needs for different techniques to meet the requirements for exploring grid semantics efficiently. As an example, we used a specific grid problem related to the selection of application variants in a domain and propose the use of a grid ontology and a reasoning approach to address it.
Autonomous Scheduling through Self-Organizing Computational Overlay Network in Decentralized Desktop Grid Systems
HongSoo Kim, EunJoung Byun, ChanYeol Park, MaengSoon Baik, ChongSun Hwang, Korea University, Republic of Korea.
  Abstract: This paper addresses the organization of computation environment and the scheduling strategies to achieve reliable execution and high throughput in decentralized desktop grid systems. Desktop grid systems can be characterized by heterogeneous, volatility, scalability, and extreme dynamism of their computation environment. In the environment, the organization of dependable computation environment and efficient scheduling schemes pose significant challenges. However, most current approaches are based on centralized computing environment. Accordingly, these approaches suffer from a low scalability and an overhead of the central server. Recently, decentralized approaches have been employed, but do not guarantee a reliable computation and a high throughput due to a lack for the supports of dynamic environments and efficient scheduling. Therefore, in this paper a Self-Organizing Computational Overlay Network (SOCON) through coordinator and volunteer overlay network is proposed to achieve reliable computation environment in decentralized way. Moreover, an A-FCFS (Autonomous First-Come First-Served) scheduling on the SOCON reflecting dynamic volunteer characteristics is proposed to further improve performance through adaptive behavior of agents. Performance evaluation demonstrates that the proposed A-FCFS scheduling on the SOCON improves throughput using Korea@Home.
On the Improvement of the Network QoS in a Grid Environment
Agustín Caminero Herráez, Carmen Carrión Espinosa, Blanca Caminero Herráez, Universidad de Castilla La Mancha, Spain.
  Abstract: Grid computing technologies are increasingly being used to aggregate computing resources that are geographically distributed across different locations. Because of the necessity to interconnect those resources, networks play an important role in grid computing, and their influence should be considered. Also, network quality of service (QoS) is a term whose importance has been growing up during last years, but this increasing importance has been almost totally skipped by the grid middleware researchers. What we want to do is to give the network the significance it deserves by means of developing a network-aware grid meta-scheduler, also known as grid network broker, or GNB. This entity would be assigned the care of utilization of the links, in order to prevent them from overloading. To achieve that, several algorithms might be used. Our GNB will be based on the Differentiated Services architecture and will allow different algorithms to be implemented within itself, so that different admission and scheduling strategies can be studied. In this paper, we discuss our grid network broker model, which is being developed using GridSim.
A Decentralized Service Discovery Algorithm for Grid Environment
Jianjun Hu, Technology Center of Software Engineering, Institute of Software, Chinese Academy of Sciences, P.R. China
Heqing Guan, Hua Zhong, Technology Center of Software Engineering,Institute of Software, Chinese Academy of Sciences, P.R. China.
  Abstract: A computational grid is an emerging computing infrastructure that enables effective access to high performance computing resources. Service discovery is one of the most important research issues of grid computing. However, traditional service discovery algorithms are not suitable for grid environment due to the facts of inherited decentralization and loose coupling of grid application itself. To address this problem, we in the paper analyze the limitation of those traditional algorithms, and present a novel grid oriented service discovery algorithm, namely DSDA, which is essentially decentralized. DSDA also takes network latency and processing capability of service nodes into account. Our later experiment justified DSDA''s effectiveness in the grid environment.
Organization of Grid Resources in Communities
Asif Akram, CCLRC Daresbury, Great Britain
Robert Allan, CCLRC, Great Britain.
  Abstract: Locating suitable resources within a Grid is a computationally intensive process, with no guarantee of quality and suitability of the discovered resources. An alternative approach is a middleware to categorize resources based on the services they provide leading to the interaction of peers with common goals to form societies/communities. The middleware organizing resources in different communities is suggested to be useful for efficient resource discovery. The communities can be adaptive in nature and evolve based on changes in their operating environment -- such as changes in neighboring communities and user requirements. We have implemented JXTA prototype to illustrate the concepts of community formation in which Peers offering different services can be grouped together based on different criteria.
Interoperable Grid Agent Middleware for Scientific Instruments and Sensors
A.B.M. Russel, Asad I. Khan, David Abramson, Monash University, Australia.
  Abstract:Service oriented computational Grid requires interoperable Grid agent middleware in order to interact with scientific instruments and sensors. Robust interaction among service oriented Grid applications and instruments or sensors can be achieved by a set of message sequences forming a communication protocol on demand by a Grid agent middleware. A Grid agent wishing to interact with a large number of heterogeneous instruments or sensors that can get modified in times does not need to have detailed knowledge of how to interact with any instruments or sensors services. This requires a Grid agent implement the communication protocol for the instrument or sensors on the fly to interact using application-specific methods of a Grid application with scientific instruments or sensors. This paper presents such a Grid agent framework that can leverage the cost of changing any Grid application when the underlying scientific instruments or sensors are updated.