Xiaoshe Dong

A constellation model for grid resource management

By analyzing the demand of the Grid resource management, this paper proposes a constellation model for dynamically organizing and managing Grid resources according to the integrated service capabilities of each node. The logical layer of the constellation model matches with the underlying physical organization. Some evaluation criterions for the integrated service capabilities are proposed in this paper, which can also be used as the criterion in selection of the standby fixed star node in resource management. By defining the minimum resource management unit, the model is more suitable to manage the resources dynamically and easier to realize uniform resource management. The services in the constellation model can be developed according to WSRF and the model conforms to OGSA.

Key techniques of software sharing for on demand service-oriented computing

In this paper a software sharing system is developed in the grid environment to enforce On Demand Computing policy and maximize the usage of both hardware and software resources. The system adopts the constellation model for resource management and combines the sharing and scheduling of both hardware and software license resources to address the ever-increasing demands of software resource sharing in Grid and for Service-oriented computing. The systems ability to sustain the softwares legacy GUI helps reduce the complexity of system integration and enhance usability.

A Scalable and Low-Cost Grid Portal

By analyzing various current grid portal designing modules, we propose pluggable based-on-OSGi grid portal, called BOPGP, to guarantee grid portal independent of platforms. BOPGP is designed based on OSGi, encapsulating grid service into dynamically installed and uninstalled Bundle, which can enlarge and modify system functions. To better satisfy the users requirement of on-demand computing, we also combine the asynchronous and synchronous request technologies to decrease grid portals response time, and give some formalized descriptions of co-operation between users and portal, to measure the spending of users computing through portal. Parameters such as operating cost, operating utility, are defined to evaluate the portal performance. Consequently, we put BOPGP into software sharing system. Analyzing experimental results, it is proved that BOPGP has fine scalability and low cost.

A high availability mechanism for parallel file system

Parallel file systems achieve a high I/O throughput by dividing a file into multiple blocks and storing them on multiple I/O nodes. However, the reliability and availability of the parallel file systems are sacrificed for the stripping of file data over multi I/O nodes. A new mechanism named Logic Mirror Ring (LMR), has been developed to improve the reliability and availability of the parallel file systems in this study. A logic mirror ring is built over all I/O nodes to indicate the mirror relationship among the nodes, i.e., each node maintains not only its own data but also the mirror data of other nodes. The fault tolerant capability of the system is improved because the node maintaining the mirror data of the failed node will take over the requests to the failed node. The mirror depth can be adjusted to different levels based on the requirements of the reliability and availability. A model is developed to evaluate the reliability and availability of the parallel file systems. The effects of LMR on the reliability and availability of the parallel file system is studied. The results show that LMR can be used to improve the reliability and availability of the parallel file systems effectively.

mPlogP: A Parallel Computation Model for Heterogeneous Multi-core Computer

Due to the heterogeneity and the multigrain parallelism of the heterogeneous multi-core computer, communication and memory access show hierarchical characteristics ignored by other models. In this paper, a new model named mPlogP, is presented on the basis of the PlogP model, in which communication and memory access is abstracted by considering these new characteristics of the heterogeneous multi-core computer. It uses memory access to model the behavior of computation, estimates the execution time of every part of applications and guides the optimization of effective parallel programs. Finally this proposed model is validated by experiments that it can precisely evaluate the execution of parallel applications under the heterogeneous multi-core computer.

Fuzzy Petri Net-Based Optimized Semantic Web Service Composition

The rapid increasing of similar Web services in the Internet, as well as enhancements of requirements for quality of service (QoS) by users, had created great challenges for Web service composition. We propose an optimized method for Web service composition on the basis of semantic Web. Web service is described by fuzzy Horn clause, and a QoS-oriented Web service composition model is built by fuzzy Petri Net. Based on the proposed model and semantic Web, an optimized Web service composition algorithm is introduced. An example of Web service composition is presented to illustrate that the proposed model is beneficial to both building and execution of web service composition.

A Novel Software Deployment Method Based on Installation Packages

The numbers of deployed computational servers in datacenters are becoming larger and larger driven by growing processing requirements, which poses a new challenge to software deployment due to the large amount of nodes involved in the deployment process. Consequently, traditional approach software deployment can not scale well because of their time-consumption properties. In order to solve this problem and provide an efficient solution, this paper proposes a new method for deploying software automatically in large-scale clusters. Softwares are classified into categories according to their installation process. We developed a software prototype SDS: Software Deployment System to verify and test the proposed method. By improving the deployment program based on the traditional installation package, the approach can resolve the difficulty of interaction between administrator and deployment processes during software deployment to achieve the automatic deployment of software. Experimental results show that the system can efficiently deploy software on Linux cluster and outperform linear growth software deployment system.

Towards Automated Trust Negotiation for Grids

Trust is an important aspect of decision making for grid applications. It has been a fundamental but challenging problem to gain assurance of the trustworthiness of service providers or requesters. This paper proposes a novel trust negotiation framework, ENVOY, to establish trust relationship between service providers and requesters in grids. ENVOY supports various kinds of delegation by leveraging attribute-based credentials and defines delegation range expediently by using attribute constraint, and we develop a negotiation strategy based on protection tree to support ENVOY, which provides the support for protection of sensitive information of the two negotiation parties. Moreover, ENVOY employs multithreaded technology and trust ticket to speed up negotiation. This paper describes the implementation of ENVOY and designs experiments to evaluate its performance, and the experimental results show that it is applicable.

A Parameter Dynamic-Tuning Scheduling Algorithm Based on History in Heterogeneous Environments

In MapReduce model, the job execution time was prolonged by the straggler tasks in heterogeneity environments. The LATE scheduler has introduced the longest remaining time strategy, but it also has some drawbacks such as inaccurate estimated time and the wasting of system resources. In order to solve these problems, we propose two main algorithms : The parameter dynamic-tuning algorithm based history estimates progress of a task accurately since it dynamically tunes the weight of each phase of a map task and a reduce task according to the historical values of the weights, The evaluation-scheduling algorithm reduce the wasting of system resources by evaluating the free slot before launching a straggler task on this node. The two main algorithms are implemented in hadoop 0.20.1. The environment results are satisfaction to our expects and significantly reduce the wasting of system resources.

A Profile-based Memory Access Optimizing Technology on CBE Architecture

In the paper, we investigate the memory access technology on cell broadband engine architecture (CBEA), and develop a profiling infrastructure for memory management on the architecture. By registering the dynamic memory allocation and providing details of trace of memory access, the infrastructure provides the data partition information automatically which alleviates the burdens of programmer and provides a safety guarantee for aggressive data prefetch for computing task. On the other hand, the profile information is useful for analyzing the patterns of memory access and helpful for further performance optimization. Experimental results show that applications implemented based on our SDK library not only support aggressive memory access method without the requirement of external data partition information, but also could be optimized aggressively under the guideline of the profile information provided by the proposed SDK library.