Zhili Cheng

Controversy corner: Optimized QoS-aware replica placement heuristics and applications in astronomy data grid

This paper studies the Quality-of-Service (QoS)-aware replica placement problem in a general graph model. Since the problem was proved NP-hard, heuristic algorithms are the current solutions to the problem. However, these algorithms cannot always find the effective replica placement strategy. We propose two algorithms that can obtain better results within the given time period. The first algorithm is called Cover Distance algorithm, which is based on the Greedy Cover algorithm. The second algorithm is an optimized genetic algorithm, in which we use random heuristic algorithms to generate initial population to avoid enormous useless searching. Then, the 0-Greedy-Delete algorithm is used to optimize the genetic algorithm solutions. According to the performance evaluation, our Cover Distance algorithm can obtain relatively better solution in time critical scenarios. Whereas, the optimized genetic algorithm is better when the replica cost is of higher priority than algorithm execution time. The QoS-aware data replication heuristic algorithms are applied into the data distribution service of an astronomy data grid pipeline prototype, and the operation process is studied in detail.

SOAVM: A Service-Oriented Virtualization Management System with Automated Configuration

Virtualization technologies are becoming more and more widely used in the distributed computing area. In order to organize and manage various virtual machines in the SOA (Service Oriented Architecture) environment more effectively, a virtualization management system called SOAVM is proposed and demonstrated in this paper. SOAVM addresses the limitation of virtual appliances in software quantities by employing ACS (Automated Configuration Service). ACS provides a more flexible way to conduct automated software configuration on virtual machines compared to the traditional virtual appliance pattern. The system framework is designed, which defines all functions of SOAVM as services to comply with SOA. Then the system design is described, illustrating how ACS works. Furthermore, critical implementation issues are discussed in detail, which demonstrates that SOAVM is a feasible and efficient virtualization management system.

A Pipeline Virtual Service Pre-Scheduling Pattern and its Application in Astronomy Data Processing

Based on Open Grid Services Architecture (OGSA), the concept and the formal model of Pipeline Virtual Service (PVS) are proposed and presented in this paper. PVS is used to model a special type of grid workflow composed of a group of services that can be executed as a pipeline. A PVS Pre-Scheduling Pattern is described in detail, which can overlap the execution time of a PVS with the time of generating and optimizing the scheduling solution pool. The Pattern includes two independent components, the solution generating component (GenFunc) and the solution execution component (ExeFunc). GenFunc can take advantage of the powerful computing potential of grid environment to find the optimal or a near optimal scheduling solution, while ExeFunc can select a suitable scheduling solution and put a PVS into running as soon as possible. Two corresponding algorithms, SODGen and SODExe, are also developed. The analytic and simulation results show that the PVS Pre-Scheduling Pattern can not only balance the scheduling cost and scheduling precision, save the scheduling time greatly, but also reduce the execution time through selecting the optimized solution. A prototype is designed and implemented for a large scale astronomy data processing center. The early engineering evaluation experiments show that the PVS Pre-Scheduling Pattern is feasible, efficient, and flexible in real astronomy data processing.Based on Open Grid Services Architecture (OGSA), the concept and the formal model of Pipeline Virtual Service (PVS) are proposed and presented in this paper. PVS is used to model a special type of grid workflow composed of a group of services that can be executed as a pipeline. A PVS Pre-Scheduling Pattern is described in detail, which can overlap the execution time of a PVS with the time of generating and optimizing the scheduling solution pool. The Pattern includes two independent components, the solution generating component (GenFunc) and the solution execution component (ExeFunc). GenFunc can take advantage of the powerful computing potential of grid environment to find the optimal or a near optimal scheduling solution, while ExeFunc can select a suitable scheduling solution and put a PVS into running as soon as possible. Two corresponding algorithms, SODGen and SODExe, are also developed. The analytic and simulation results show that the PVS Pre-Scheduling Pattern can not only balance the scheduling cost and scheduling precision, save the scheduling time greatly, but also reduce the execution time through selecting the optimized solution. A prototype is designed and implemented for a large scale astronomy data processing center. The early engineering evaluation experiments show that the PVS Pre-Scheduling Pattern is feasible, efficient, and flexible in real astronomy data processing.

A Service Level QoS Mechanism and Algorithm for Data Distribution and Backup in an Grid Based Astronomy Data Management System

Based on the research of SLA, SNAP, MetaService and contractual computing service, a service level QoS mechanism and algorithm for astronomy data distribution and backup in grid environment is proposed. A weight based comparison method for different data services provided by data servers is described first. The method can choose suitable data servers under the requirements of consumers. Based on the weight based comparison method, a data distribution algorithm is designed and this algorithm can obtain an optimized data allocating solution with a time complexity of O(x lg x + nxy) . In a practical example of data distribution and backup, the balancing mechanism is explained between two factors: the performance and the cost. The mechanism and the algorithm have been implemented in an astronomy data management prototype system. Compared to non- QoS guarantee scenario, a set of experiments with different weights of performance and cost have been tested and the results show that the QoS guarantee method can satisfy consumers requirement effectively.

Virtual Workspace Based Job Execution Methods and Dynamic Makespan Optimized Scheduling Algorithms

The virtualization technology is helpful to support grid QoS. A manual configuration method and an automatic configuration method are designed to build the virtual workspace environment based on virtualization technology. Both the MetaServices and the automatic configuration software are employed to establish the virtual workspace environment. For dynamic jobs, the corresponding dynamic makespan optimized (DMO) scheduling algorithms are developed. Results of the simulation experiments demonstrate that, compared to models without virtualization support, the job execution methods and DMO scheduling algorithms proposed in this paper can achieve higher throughput, optimize job makespan, and improve the QoS of grid systems.

A Grid QoS Provision Solution for PGAS based on MetaService and CCM

This paper mainly focuses on solution to QoS provision in power grid analysis system (PGAS) which is an improvement for the electric power information system. The solution consists of two parts. The first part is the grid QoS provision model for PGAS based on metaservices and CCM. The model is proposed based on requirement of data-intensive QoS guaranteed grid systems. The second part is an adaptive scheduling algorithm which handles two types of tasks (with/without QoS guarantee tasks) in PGAS. Simulations of two types of tasks based on GridSim show the results of our solution can meet various QoS requirements of users in PGAS and the scheduling algorithm take good effect.

Optimizing Service Data Transmission in an OGSA Based Grid System

The problem of providing suitable data transmission mechanism among services is critical in an Open Grid Services Architecture (OGSA) based grid system. Four typical data transmission methods between two services are provided first. Then for one complex data transmission patterns among multiple services, the corresponding data transmission scheduling mechanism is discussed in detail. Based on the steps of one data transmission path between two services, a data transmission performance prediction model is described. Together with some implementation and environment dependent parameters, the model can provide different performance prediction functions for different data transmission methods. At the same time, a set of comparison experiments have been designed and tested in a grid environment built up with Globus Toolkit 3. The comparison results show that the theoretical data transmission performance is very close to the experimental performance. At the end of the paper, a grid data transmission optimizing MetaService is designed to select suitable transmission mechanism in a grid environment dynamically. The MetaService can adapt to different scenarios, and improve the data transmission performance.