Chuliang Weng

The hybrid scheduling framework for virtual machine systems

The virtualization technology makes it feasible that multiple guest operating systems run on a single physical machine. It is the virtual machine monitor that dynamically maps the virtual CPU of virtual machines to physical CPUs according to the scheduling strategy. The scheduling strategy in Xen schedules virtual CPUs of a virtual machines asynchronously while guarantees the proportion of the CPU time corresponding to its weight, maximizing the throughput of the system. However, this scheduling strategy may deteriorate the performance when the virtual machine is used to execute the concurrent applications such as parallel programs or multithreaded programs. In this paper, we analyze the CPU scheduling problem in the virtual machine monitor theoretically, and the result is that the asynchronous CPU scheduling strategy will waste considerable physical CPU time when the system workload is the concurrent application. Then, we present a hybrid scheduling framework for the CPU scheduling in the virtual machine monitor. There are two types of virtual machines in the system: the high-throughput type and the concurrent type. The virtual machine can be set as the concurrent type when the majority of its workload is concurrent applications in order to reduce the cost of synchronization. Otherwise, it is set as the high-throughput type as the default. Moreover, we implement the hybrid scheduling framework based on Xen, and we will give a description of our implementation in details. At last, we test the performance of the presented scheduling framework and strategy based on the multi-core platform, and the experiment result indicates that the scheduling framework and strategy is feasible to improve the performance of the virtual machine system.

Dynamic adaptive scheduling for virtual machines

With multi-core processors becoming popular, exploiting their computational potential becomes an urgent matter. The functionality of multiple standalone computer systems can be aggregated into a single hardware computer by virtualization, giving efficient usage of the hardware and decreased cost for power. Some principles of operating systems can be applied directly to virtual machine systems, however virtualization disrupts the basis of spinlock synchronization in the guest operating system, which results in performance degradation of concurrent workloads such as parallel programs or multi-threaded programs in virtual machines. Eliminating this negative influence of virtualization on synchronization seems to be a non-trivial challenge, especially for concurrent workloads. In this work, we first demonstrate with parallel benchmarks that virtualization can cause long waiting times for spinlock synchronization in the guest operating system, resulting in performance degradation of parallel programs in the virtualized system. Then we propose an adaptive dynamic coscheduling approach to mitigate the performance degradation of concurrent workloads running in virtual machines, while keeping the performance of non-concurrent workloads. For this purpose, we build an adaptive scheduling framework with a series of algorithms to dynamically detect the occurrence of spinlocks with long waiting times, and determine and execute coscheduling of virtual CPUs on physical CPUs in the virtual machine monitor. We have implemented a prototype (ASMan) based on Xen and Linux. Experiments show that ASMan achieves better performance for concurrent workloads, while maintaining the performance for non-concurrent workloads. ASMan coscheduling depends directly on the dynamic behavior of virtual CPUs, unlike other approaches which depend on static properties of workloads and manual setting of rules. Therefore, ASMan achieves a better trade-off between coscheduling and non-coscheduling in the virtual machine monitor, and is an effective solution to this open issue.

An In-VM Measuring Framework for Increasing Virtual Machine Security in Clouds

Cloud computing relies heavily on virtualization. Virtualization technology has developed rapidly because of the rapid decrease in hardware cost and concurrent increase in hardware computing power. A virtual machine monitor (VMM, also called a hγpervisor) between the hardware and the OS enables multiple virtual machines (VMs) to run on top of a single physical machine. The VMM manages scheduling and dispatching the physical resources to the individual VMs as needed, and the VMs appear to users as separate computers. Widely used virtualization technologies include VMWare, Xen, Denali, and the Kernel-Based Virtual Machine (KVM). In this framework, a module measures executables running in virtual machines (VMs) and transfers the values to a trusted VM. Comparing those values to a reference table containing the trusted measurement values of running executables verifies the executable/s status.

Automatic Performance Tuning for the Virtualized Cluster System

System virtualization can aggregate the functionality of multiple standalone computer systems into a single hardware computer. It is significant to virtualize the computing nodes with multi-core processors in the cluster system, in order to promote the usage of the hardware while decrease the cost of the power. In the virtualized cluster system, multiple virtual machines are running on a computing node. However, it is a challenging issue to automatically balance the workload in virtual machines on each physical computing node, which is different from the traditional cluster systems load balance. In this paper, we propose a management framework for the virtualized cluster system, and present an automatic performance tuning strategy to balance the workload in the virtualized cluster system. We implement a working prototype of the management framework (VEMan) based on Xen, and test the performance of the tuning strategy on a virtualized heterogeneous cluster system. The experimental result indicates that the management framework and tuning strategy are feasible to improve the performance of the virtualized cluster system.

An economic-based resource management framework in the grid context

The economic mechanism is very suitable for solving the problem of resource management in the grid computing environment, and it can guarantee the interest of participators in the grid with fairness and efficiency. In this paper, we propose an economics-based resource management framework for grid computing, and then we focus on how to determine the price of resources with the economic mechanism. We present a general equilibrium method for general resources and a double auction method for special resources in the grid environment respectively. Simulations are performed and the experimental results indicate that the two methods are effective for corresponding application scenarios.

An Online Scheduling Algorithm for Assigning Jobs in the Computational Grid

The computational grid provides a promising platform for the deployment of various high-performance computing applications. Problem in implementing computational grid environments is how to effectively use various resources in the system, such as CPU cycle, memory, communication network, and data storage. There are many effective heuristic algorithms for scheduling in the computational grid, however most scheduling strategies have no theoretical guarantee at all. In this paper, a cost-based online scheduling algorithm is presented for job assignment in the grid environment with theoretical guarantee. Firstly, a scheduling framework is described, where the grid environment is characterized, and the online job model is defined. Secondly, the cost-based online scheduling algorithm is presented where costs of resources are exponential functions of their loads, and the performance of this algorithm is theoretically analyzed against the performance of the optimal offline algorithm. Finally, we implement the algorithm in the grid simulation environment, and compare the performance of the presented algorithm with the other three algorithms, and experimental results indicate that the cost-based online scheduling algorithm can outperform the other three online algorithms.

A double auction mechanism for resource allocation on grid computing systems

Considering dynamic, heterogeneous and autonomous characteristics of computing resources in grid computing systems and the flexibility and effectivity of economics methods applied to solve the problem of resource management, a double auction mechanism for resource allocation on grid computing systems is presented. Firstly, a market model of double auction is described, in which agents are utilized to represent the computational resource traders in the grid environment and are equipped with the reinforcement learning algorithm. Secondly, a double auction mechanism is presented, where the uniform-price auction is adopted aiming at CPU resources, and the transaction fee can be adjusted flexibly. Finally, the efficiency of the presented double auction mechanism is analyzed through experiments, and experimental results show that the presented mechanism is efficient, and the transaction price varies mildly.

A QoS-Aware and Fault-Tolerant Workflow Composition for Grid

The continuous increasing of computing power in electric system computing places a threshold to the single host use and suggests an approach based on distributed computing. An emerging solution is grid technology, which allows organization to make better use of existing computing resources by providing them with a single, transparent, aggregated source of computing power. New generation of grid infrastructure, where Web services are building blocks, allow management of a Web services workflow. In this paper, a novel strategy - QoS-aware and fault-tolerant workflow composition - has been applied to utilize Web services and business process execution language for overcoming the issues about task assignment, security, flexibility and workflow management. This new strategy is lightweight Web services based computing power-sharing architecture, and it is not only suitable for executing computing works which are able to run in batches, but also be able to solve current issues in Web services based computing application such as system resilience, fault tolerance, efficiency of job scheduling and the instability in congested network environment.

Data Aggregation with Multiple Spanning Trees in Wireless Sensor Networks

In wireless sensor networks, the data aggregation is an essential paradigm for routing, through which the multiple data from different sensors can be aggregated into a single data at intermedial nodes enroute, in order to eliminate data redundancy and achieve the goal of saving energy. Some existed medium access protocols and algorithms can effectively prolong the lifetime of the sensor network by determining when each sensor should transmit its data, and when it should sleep. In this paper, we focus on applying multiple spanning trees to organize the data aggregation, which is different from these existed single spanning tree methods. At first, the problem of constructing multiple spanning trees is transformed into a linear programming problem of the data flow network. Based on the solved optimal rate between the two adjacent sensors, the two constructing algorithms of the spanning tree are presented. Experimental results indicate that the method of multiple spanning trees can be of benefit to energy saving for wireless sensor networks, and the corresponding appropriate constructing algorithm can prolong the lifetime of the sensor network.

A Distributed Approach for Resource Pricing in Grid Environments

A distributed group-pricing algorithm is presented for the market-based resource management in the grid context based on quick convergence of centralized algorithms and scalability of distributed algorithms. According to the new algorithm, resources in the system are divided into multiple resource groups according to the degree of price correlation of resources. When the demand and supply of resources in the system changes, each auctioneer in the defined system structure is responsible for adjusting simultaneously the price of one resource group respectively until the excess demand of all resources becomes zero. We test the distributed group-pricing algorithm against the existed distributed algorithm, and analyze the property of the algorithm. Experimental results indicate that an equilibrium can be achieved by the distributed group-pricing algorithm quicker than by the existed distributed algorithm.