Conference Paper

Modeling and Performance Analysis of a Priority-based Scheduling Scheme in Service Grid

State Key Lab. of Networking & Switching Technol., Beijing Univ. of Posts & Telecommun.
DOI: 10.1109/GCC.2006.61 Conference: Grid and Cooperative Computing, 2006. GCC 2006. Fifth International Conference
Source: IEEE Xplore


A multiserver multiqueue (MSMQ) based scheduling system of the service grid system and a modeling and analysis method based on stochastic Petri net (SPN) are proposed. Due to the cost and resources consumed are different for difference grid service requests, the multiple priority queues are designed in the grid scheduler of the scheduling system. In order to promote the throughput, we propose a priority-based advanced shortest expected delay first scheduling algorithm (P-ASED) which consists of two steps: the first step is request selecting according to the priority of requests; and the second step is request dispatching according to the weighted average of the expected delay and relative queue length of service nodes. The simulation experiments verify that the proposed MSMQ-based scheduling system and the P-ASED algorithm can schedule the service request according to the priority and promote the throughput of the service grid system

0 Reads
  • [Show abstract] [Hide abstract]
    ABSTRACT: Reasonable and fault-tolerance scheduling mechanism is the key factor of effective use the grid resource and guarantee the implement of applicable task. For Grid-based Distributed Control System (GDCS), this paper introduces the two-level task scheduling system frame: System scheduler sorts all upload tasks by functions or attributes, then assigns local area scheduler to dispatch tasks to resource nodes, and handles fault-tolerance dynamically during the process of task. Considering two types subtask of complete independence and non complete independence of GDCS, establishes a fault-tolerance mechanism model by Stochastic Petri Nets (SPN) for local areas scheduler, describes the complete independent subtask dispatching and allocating policies, and proposes a dynamic scheduling train of thought of non complete independent subtask.
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The computational grid is a new technology aimed to facilitate resource sharing and wide area collaborations. Task scheduling is an important aspect in the grid environment to obtain a desired quality of service (QoS). In grid environments, the resource management system (RMS) is responsible for the scheduling and execution of tasks. The RMS divides the incoming tasks into several subtasks and distributes them among grid resources for parallel execution. Then, the RMS integrates the received results into the entire task output, which is requested by the user. A grid service is desired to execute a certain task under the control of the RMS. In this paper, the task scheduling by RMS and the task execution within grid resources are modeled using coloured Petri nets (CPNs). The proposed CPN-based modelling pattern formally describes the process of task distribution and execution within the grid environment. We have also proposed a method for evaluation the grid service reliability based on the analysis of the model. In addition, an instance of the proposed model for a sample grid environment is constructed and analysed using CPN Tools. The CPN model and the results of the reliability evaluation of the sample grid environment are also presented in this paper.
    Future Generation Computer Systems 10/2010; 26(8-26):1141-1150. DOI:10.1016/j.future.2010.05.015 · 2.79 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, a mathematical solution for reducing the mean number of the tasks waiting to be processed in a grid environment is proposed. The approach uses queuing theory and models the grid environment in the form of an open queuing network (QN). Applying the steady state analysis to the proposed QN and minimizing the mean number of waiting tasks within the grid environment, an equality and inequality system is obtained. Solving the equality and inequality system, subtasks arrival rates at each of the resources within the grid environment can be estimated. Applying the obtained subtasks arrival rates at each of the grid resources, the mean number of the waiting tasks in the grid environment could be minimized. To evaluate the results obtained from the proposed QN and provide a graphical representation of the grid environment a formal description of the environment in terms of generalized stochastic Petri nets (GSPNs) is presented. Steady state analyzing of the GSPN model and finding the subtask dispatching weights at each of the grid resources, subtask arrival rates can be estimated. Comparing the results obtained from two proposed approaches shows that the subtasks arrival rates achieved from QNs and GSPNs are the same.
    The Journal of Supercomputing 01/2012; 59(1):469-485. DOI:10.1007/s11227-010-0448-5 · 0.86 Impact Factor