GCA 2007

PREFACE.

WORKSHOP COMMITTEES.

CONTENTS.

Advance Reservation is one possible way to enable resource co-allocation on the Grid. This method requires all the resources to have advance reservation capability as well as coordination protocol support. We employed two-phase commit protocol as a coordination protocol, which is common in the distributed transaction area, and implemented an Advance Reservation Manager called PluS. PluS works with existing local queuing managers, such as TORQUE or Grid Engine, and provides users advance reservation capability. To provide the capability, there are two implementation methods; 1) completely replaces the scheduling module of the queuing manger, 2) represents reservation as a queue and controls the queues using external interface. We designed and implemented a reservation manager with both way, and evaluated them. We found that the former has smaller overhead and allows arbitrary scheduling policy, while the latter is much easier to implement with acceptable response time.

Desktop Grid is usually used to execute a lot of user jobs at dispersed sites. However, how to trust a grid site and where to schedule the user jobs is a key factor that affects success rate, security, and fault tolerance of the job scheduling. A computational model of reputation in desktop grid is proposed and the user jobs are always scheduled to sites with just high reputation values. In order to avoid reputation-competing problem (i.e., all jobs scheduled to the site with the highest reputation values), we introduce the reputation values balancing concept similar to load balancing. A reputation balancing algorithm is proposed which aims to minimize the deviation of reputation values and the number of jobs scheduled to a grid site is proportional to the site's reputation value. Simulation results on RSBSME (Remote Sensing Based Soil Moisture Extraction) workload in a real desktop grid environment show that the performance of the proposed reputation model is effective, fault tolerant and scalable.

This paper presents the design, implementation and evaluation of a dataflow system, including a dataflow programming model and a dataflow engine, for coarse-grained distributed data intensive applications. The dataflow programming model provides users with a transparent interface for application programming and execution management in a parallel and distributed computing environment. The dataflow engine dispatches the tasks onto candidate distributed computing resources in the system, and manages failures and load balancing problems in a transparent manner. The system has been implemented over .NET platform and deployed in a Windows Desktop Grid. This paper uses two benchmarks to demonstrate the scalability and fault tolerance properties of our system.

Gfarm was designed and implemented for achieving high scalability of total I/O bandwidth for a comparatively small number of files that have a large file size. In this paper, the performance of Gfarm version 1.4 is measured by both metadata-intensive and data-intensive benchmarks, in order to analyze Gfarm's characteristics in terms of granularity of file operations and file size. The results show a conceivable bottleneck in the use of Gfarm as a cluster filesystem, hints for parameter tuning, and types of applications suitable for use with Gfarm. In our experiment, 5600 files with a file size of 5MB, on average, were created over 112 nodes and read in turn, and a total of 2.2 [GB/sec] for write and a total of 4.2 [GB/sec] for read throughput were achieved.

Workflow scheduling is a very important system function that Grid systems have to support. But the scheduling of workflow tasks is an NP complete problem. In this paper, we propose a new scheduling method– “temporal decomposition” – which first divides a whole grid workflow into some sub-workflows and then schedules them using a new efficient algorithm. By dividing a large problem (workflow) into smaller problems (sub-workflows), the “temporal decomposition” can achieve much lower computation complexity. Numerical results show that our proposed scheme is more efficient in comparison with a well known existing grid workflow scheduling method.

The Grid seeks to provide a universal platform for collaborations among organizations to merge their resources together in a reliable and scalable manner for solving problems that are multidisciplinary in nature. This paper presents a model for representing a problem as a flow based on the semantics of WS–BPEL. In particular, the model permits a link to be declared inside not only a parallel task but also a sequential task. In addition, the concept of normal forms is presented to eliminate ambiguity in the model by regulating the use of links. Several comparisons are also made with a mix of literatures and standards to assess the efficacy of the model itself.

Job workflow execution can be classified into centralized execution and distributed execution. The centralized execution may cause problems such as single point of failure and poor scalability; whereas the distributed job workflow execution may bring additional runtime overhead. This paper describes a comparison study between these two execution models. First, to put discussion into prospective, our framework for mobile agent based distributed job workflow execution over the Grid: Mobile Code Collaboration Framework (MCCF) is described. Second we introduce a commonly used centralized execution engine, Condor DAGMan. Then, the differences between job workflow execution in our context and those in the Condor DAGMan are discussed. Finally, a comparison study using simulated job workflows executed on a prototype implementation of the MCCF and Condor DAGMan is carried out on an emulated WAN setup. The results show that MCCF achieves better job workflow execution time.

Rendering of images is a very compute intensive task. Thus, it was chosen as one of the prospective commercial market that could leverage on grid/cluster technology. This paper reports on the development and deployment of grid rendering service across a heterogeneous grid environment. It covers the entire process from the submission of the jobs to management and rendering of the model. The prototype was successfully deployed and the results show the feasibility as well as the advantage of using the Grid in rendering animation.

Grid computing has evolved dramatically, migrating to service oriented Grids: the third generation Grids. As a result, there has been great interest from both industry and the research community in enabling collaborative service provisioning through operational virtual communities over the grid. However, the existing service providing mechanism of the Grid is too rigid to provide the flexibility for a wide range of collaborative services. Lacking virtual community support at the operation level becomes a major barrier to promoting collaborative services over the Grid environment. In this paper we propose a collaborative workspace over service-oriented grid for developing operationally transparent virtual communities in a wide variety of domains.

This paper describes a new approach (Opal OP: Opal Operation Provider) to wrap existing legacy applications as Grid services. In order to expose, with minimal effort, existing applications as Grid services, Opal OP provides a method for wrapping a legacy application as a program module, or as an operation provider. Traditional wrapping methods usually restrict the way to implement Grid services because these methods provide only a suite of interfaces necessary for using the wrapped application. The proposed Opal OP, on the other hand, doesn't restrict the way to implement a Grid service from a legacy application. Opal OP is imple-mented as an operation provider that wraps the application and thus can be used as a module in the Grid service. Application developers can easily develop their own services where legacy applications are wrapped through the utilization of Opal OP. In this paper, we show some scientific applications including a bio-molecular simulation system developed as Grid services using Opal OP. The results show the usefulness and effectiveness of Opal OP.

As the prevalence and coverage of grid deployment gather momentum, new tools are needed to support its effective operation and debugging. In this paper, we will introduce the design and implementation of our grid security monitoring tool, which enables multi-organization grid operation center to detect and visualize security abnormalities. Security sensors are deployed across the grid to collect security-related information from globus gatekeeper and jobmanager and store the information in a central database. The information is then visualized in an intelligent manner for security error/violation detection and resolution. The tool was successfully deployed and used across the Pacific Rim Application and Grid Middleware Association (PRAGMA) grid test-bed, where a variety of authentication and authorization failures were quickly detected and resolved. To ensure that the tool would be part of a set of tools for deployment at the Grid Operation Center, the tool was built as an extension module of the Multi-organizational Grid Accounting System (MOGAS), and shares some of the attributes used by other modules.

Drawing intermediate frames, so called inbetweens, is a very time consuming and labor-intensive task in the production of cartoon animation. Many approaches have been investigated to automate the inbetweening process. In this paper we propose a distributed rendering solution for inbetweening based on our novel modeling method, Disk B-spline. Curve The experimental results show that, compared to a single workstation, the cluster based framework provides much better performance with more than 85% of time being saved. Moving forward, we believe that the Grid will be an ideal solution to improve the animation productivity further in the future.

Comparative genomics provides a powerful tool for studying evolutionary changes among organisms. We have designed a new tool, called Phenotype Genotype Explorer to nominate candidate genes responsible for a given phenotype. There are huge datasets involved which makes this approach impractical on traditional computer architectures leading to prohibitively long runtimes. In this paper, We present a computational architecture based on a desktop grid environment and commodity graphics hardware to significantly accelerate the comparative genomics application. We present the deployment and evaluation of this approach on our grid testbed for the comparison of microbial genomes.

The Grid portal provides an environment where the user can access Grid resources, execute and monitor Grid jobs, and collaborate with other users. This paper reports on the development of the digital media Grid portal specifically to meet the needs of the media Grid community, which supports digital media processing, delivery, and storage in the Grid. The architecture of the media Grid portal and the solutions for system performance optimization and fault-tolerance are described in detail.

The PC grid initiative at NUS, which is better known as the Tera-scale Campus Grid (TCG@NUS), is the single largest PC grid on campus in the region with more than 1,400 PCs linked up today. The creation of multi TFLOPS (Trillions of Floating-point Operations per Second) of computing power to support large-scale research computation was done without additional investment on High Performance Computing (HPC) hardware, a unique feature of such grid computing implementation that will be shared in this paper. Details on the returns of investment and the results of some case studies will be presented to demonstrate the viability and the strategic value of PC grid computing as an alternative to the traditional HPC approach for research computing.

The PC grid technologies allow the PC grid infrastructure to be extended to include not just desktops but also servers and clusters to form a single pool of computation resources. It will be shown in this paper that PC grid technologies are well suited for building the high throughput computational grid. The paper will also explore the possibility of extending the campus based infrastructure beyond the organisation boundary.

The super-resolution (SR) imaging refers to the image processing algorithms for overcoming the inherent limitations of the image acquisition systems to produce high-resolution images from their low-resolution counterparts. In our recent work, a stochastic SR imaging framework has been successfully developed by applying the Markov chain Monte Carlo (MCMC) technique and shown as a promising approach for addressing the SR problem. To further overcome the intensive computation requirement of the stochastic SR imaging, Grid computing is resorted in this paper to break down the computationally-intensive MCMC SR task into a set of independent and small sub-tasks for parallel computing in the Grid computing environment. Experiments are conducted to show that Grid computing can effectively accelerating the execution time of the stochastic SR algorithm.

The existing Collaborative Virtual Environment (CVE) systems have limited scalability due to the constraints in computer processing power and network bandwidth of participating hosts. In this paper, we propose a grid-based large-scale collaborative virtual environment architecture to scale across multiple geographically dispersed resources. The architecture consists of distributed mobile agents working cooperatively in supporting and managing the collaborative virtual environments. The mobile agents are autonomous and have the ability of migrating among hosts to maximize resource utilizations. Grid technologies allow the mobile agents to execute and communicate securely in multiple administrative domains. Grid-based scheduling components and polices can be integrated to provide intelligent resource optimizations. The result will be a more scalable architecture for supporting large-scale collaborative virtual environment.

This paper describes our current work in designing an experimental gaming platform for simulating the trading of grid resources. The open platform allows researchers in grid economics to experiment with different market structures and pricing models. We would be using a design science approach in the implementation. Key design considerations and an overview of the functional design of the platform are presented and discussed.