Performance evaluation of StarPU schedulers with preconditioned conjugate gradient solver on heterogeneous (multi-CPUs/multi-GPUs) architecture

Recently fair division theory has emerged as a promising approach for the allocation of multiple computational resources among agents. While in reality agents are not all present in the system simultaneously, previous work has studied static settings where all relevant information is known upfront. Our goal is to better understand the dynamic setting. On the conceptual level, we develop a dynamic model of fair division, and propose desirable axiomatic properties for dynamic resource allocation mechanisms. On the technical level, we construct two novel mechanisms that provably satisfy some of these properties, and analyze their performance using real data. We believe that our work informs the design of superior multiagent systems, and at the same time expands the scope of fair division theory by initiating the study of dynamic and fair resource allocation mechanisms.

Under SOA, multiple services can be aggregated to create a new composite service based on some predefined workflow. The QoS of this composite service is determined by the cooperation of all these Web services. With workflow pipelining, it is unreasonable to improve the overall service performance by only considering individual services without considering the relationship among them. In this paper, we propose to allocate resources by tracing and predicting workloads dynamically with the pipelining of service requests in workflow graph. At any moment, there are a number of service requests being handled by different services. Firstly, we predict future workloads for any requests as soon as they arrive at any service in the workflow. Secondly, we allocate resources for the predicted workloads to enhance the performance by replicating more services to resources. Our target is to maximize the number of successful requests with the constraints of limited resources. Experiment shows that our dynamic resource allocation mechanism is more efficient for enhancing the global performance of composite service than static resource allocation mechanism in general.

In a shared virtual computing environment, dynamic load changes as well as different quality requirements of applications in their lifetime give rise to dynamic and various capacity demands, which results in lower resource utilization and application quality using the existing static resource allocation. Furthermore, the total required capacities of all the hosted applications in current enterprise data centers, for example, Google, may surpass the capacities of the platform. In this paper, we argue that the existing techniques by turning on or off servers with the help of virtual machine (VM) migration is not enough. Instead, finding an optimized dynamic resource allocation method to solve the problem of on-demand resource provision for VMs is the key to improve the efficiency of data centers. However, the existing dynamic resource allocation methods only focus on either the local optimization within a server or central global optimization, limiting the efficiency of data centers. We propose a two-tiered on-demand resource allocation mechanism consisting of the local and global resource allocation with feedback to provide on-demand capacities to the concurrent applications. We model the on-demand resource allocation using optimization theory. Based on the proposed dynamic resource allocation mechanism and model, we propose a set of on-demand resource allocation algorithms. Our algorithms preferentially ensure performance of critical applications named by the data center manager when resource competition arises according to the time-varying capacity demands and the quality of applications. Using Rainbow, a Xen-based prototype we implemented, we evaluate the VM-based shared platform as well as the two-tiered on-demand resource allocation mechanism and algorithms. The experimental results show that Rainbow without dynamic resource allocation (Rainbow-NDA) provides 26 to 324 percent improvements in the application performance, as well as 26 percent higher average CPU utilization than traditional service computing framework, in which applications use exclusive servers. The two-tiered on-demand resource allocation further improves performance by 9 to 16 percent for those critical applications, 75 percent of the maximum performance improvement, introducing up to 5 percent performance degradations to others, with 1 to 5 percent improvements in the resource utilization in comparison with Rainbow-NDA.

A matching Schur complement preconditioning technique for inverse source problems

Machine-to-Machine (M2M) communication is an emerging technology for communication between machines without human intervention. The main issue for cellular operators and service providers is to find a way to manage the high volume of these machine type communications and current human type communications for allocating network resources. An overall solution for large access management issues is to use application prioritization and priority class based traffic scheduling. In order to enhance the system efficiency and to reduce waste of resources, this paper investigates a creative type based dynamic priority scheduling mechanism for M2M/H2H coexistence traffics in LTE-A networks. Dynamic resources allocation is a significant challenge due to the uncertainty introduced by the network, QoS and energy efficiency requirements. The proposed solution is based on type-2 fuzzy logic controller expert system to anticipate and manage uncertainties. The performance of the proposed algorithm is evaluated by various metrics such as delay, throughput, and bandwidth utilization.

According to Internet of Things (IoT) Analytics, soon, the online devices in IoT networks will range from 25 up to 50 billion. Thus, it is expected that IoT will require more effective and efficient analysis methods than ever before with the use of Machine Learning (ML) powered by Fifth Generation (5G) networks. In this paper, we incorporate the K-means algorithm inside a 5G network infrastructure to better associate devices with Base Stations (BSs). We use multiple datasets consisting of user distribution in our area of focus and propose a Dynamic Resource Allocation (DRA) technique to learn their movement and predict the optimal position, RB usage and optimize their resource allocation. Users can experience significantly higher data rates and extended coverage with minimized interference and in fact, the DRA mechanism can mitigate the need for small cell infrastructure and prove a cost-effective solution, due to the resources transferred within the network.

Inspired by the manufacturing costs proportion of part feeding in automotive mixed-model assembly lines (MMALs) being up to 20–35%, this paper takes the dynamic scheduling of part feeding for automotive MMALs as a crucial and complex problem. Therefore, a dynamic scheduling mechanism basing on the knowledge base (KB) and fruit fly optimization algorithm (FOA) with variable step sizes and logistic chaos (VSCFOA)-enhanced general regression neural network (VSCFOA-GRNN) is proposed to tackle the real-time part feeding scheduling problem of tow trains under the dynamic manufacturing system. A mathematical model is developed to illustrate the problem, where the throughput of the assembly line and the material delivery distance are determined as components of the objective function. Subsequently, samples of the MMAL are generated by the plant simulation software and used to train the VSCFOA-GRNN model off-line. Afterward, the trained model and KB are adopted in the real-time scheduling process to determine the optimal scheduling rule combination. Finally, the effectiveness, feasibility and accuracy of the novel scheduling mechanism are validated by computational results, especially in dynamic scheduling processes. It can cope well with changes in the dynamic environment, thus effectively realizing the higher productivity of assembly lines and better system performance.

Orthogonal frequency division multiplexing (OFDM) with dynamic scheduling and resource allocation is a key component of most emerging broadband wireless access networks such as WiMAX and LTE (long term evolution) for 3GPP. However, scheduling and resource allocation in an OFDM system is complicated, especially in the uplink due to two reasons: (i) the discrete nature of subchannel assignments, and (ii) the heterogeneity of the users' subchannel conditions, individual resource constraints and application requirements. We approach this problem using a gradient-based scheduling framework. Physical layer resources (bandwidth and power) are allocated to maximize the projection onto the gradient of a total system utility function which models application-layer Quality of Service (QoS). This is formulated as a convex optimization problem and solved using a dual decomposition approach. This optimal solution has prohibitively high computational complexity but reveals guiding principles that we use to generate lower complexity sub-optimal algorithms. We analyze the complexity and compare the performance of these algorithms via extensive simulations.

In the process of transmission of digital-encoded video bitstreams over an ATM network, cells are inevitably exposed to delays and losses due to the statistical multiplexing and dynamic resource allocation mechanisms used in these networks. Error-loss concealment techniques have been proposed to limit the impact of such impairments over the video quality. These techniques alone may provide acceptable results when the cell loss probability does not exceed 10/sup -3/. For higher cell loss probabilities, these techniques must be used in conjunction with hierarchical coding schemes and packetization strategies in order to obtain satisfactory results. We propose the use of a hierarchical encoding scheme and a dynamic data-partitioning strategy for the transport of MPEG-2 video streams over ATM networks.

Admission control mechanisms are crucial for an efficient resource management and fulfillment of service level agreements. Common existing techniques perform their admission decision function for new requests, without taking into consideration the priority or guarantee commitments of particular service classes. Other multi-class based approaches are usually designed for packet-oriented differentiated service networks. This paper introduces a framework for multi-class and dynamic flow-aware admission control (MAC), which performs its admission decision by considering the class priority to which the flow requests belong and the particular resource capacity proportion allocated per-class. The main characteristic of the framework is that it performs measurements required for the load estimation and applies an admission criteria on a per-class basis, with regard to the available class-based capacity, for which a dynamic resource allocation mechanism is implemented as well.

Grid computing is an emerging paradigm to enable scalable computational environment through efficient resource utilization. In this context, resource allocation is one of the major challenges in grid environment. Currently, several approaches were proposed to solve this problem. However, most of the approaches rely on administrators' intervention for proper allocation. In this paper, we propose dynamic resource allocation mechanisms of computations using a combination of best fit algorithm and process migration.

The tremendous increase in the size and heterogeneity of supercomputers makes it very difficult to predict the performance of a scheduling algorithm. Therefore, dynamic solutions, where scheduling decisions are made at runtime have overpassed static allocation strategies. The simplicity and efficiency of dynamic schedulers such as Hadoop are a key of the success of the MapReduce framework. Dynamic schedulers such as StarPU, PaRSEC or StarSs are also developed for more constrained computations, e.g. task graphs coming from linear algebra. To make their decisions, these runtime systems make use of some static information, such as the distance of tasks to the critical path or the affinity between tasks and computing resources (CPU, GPU,...) and of dynamic information, such as where input data are actually located. In this paper, we concentrate on two elementary linear algebra kernels, namely the outer product and the matrix multiplication. For each problem, we propose several dynamic strategies that can be used at runtime and we provide an analytic study of their theoretical performance. We prove that the theoretical analysis provides very good estimate of the amount of communications induced by a dynamic strategy and can be used in order to efficiently determine thresholds used in dynamic scheduler, thus enabling to choose among them for a given problem and architecture.

In this paper, we provide simulation results for the optimal channel-aware adaptive transmission, dynamic resource allocation and multiuser scheduling schemes for the downlink of a wireless cell that is assisted by a number of relays. The relays are half-duplex and apply a regenerative decode-and-forward scheme, and the channels are Rayleigh block-faded. We apply the optimal resource allocation scheme for two specific scheduling criteria, the maximum capacity scheduler and the proportional fair scheduler. For both scheduling schemes, we study the gains of using relays on the overall aggregate throughput of the system as well as for the individual users. Furthermore, we compare the performance versus the total number of relays and their locations in the wireless cell and we study the statistics of relays' usage. Moreover, we compare between the two scheduling schemes in terms of achievable rate and fairness among users.

Network space mimicry defense constructs heterogeneous redundant configuration of execution pool through software and hardware diversification technology, and forms multi-mode output vector through scheduler. Although it can provide a point to surface fusion defense mechanism independent of traditional security means, the execution scheduler, which plays a key role, can still sniff or guess through retransmission attack because of its single and static algorithm. The basic principle of the scheduling algorithm is detected to increase the chance of mimicry escape. This paper designs a dynamic scheduling mechanism with diversified algorithms, which increases the complexity and dynamics of the scheduling mechanism, and can effectively reduce the risk of the scheduler being sniffed. Experimental results show that, compared with the traditional scheduling mechanism, the proposed diversified and dynamic scheduling scheme can significantly reduce the risk of the scheduler being attacked, and further improve the security of the mimicry defense system.

W-CDMA is the strongest candidate as the air interface technology of the third generation mobile communication systems, which are expected to support multimedia services with QoS requirements. In this paper, we review the W-CDMA technology and consider the issues in multimedia support. We propose the dynamic resource scheduling mechanism for QoS provisioning in W-CDMA through optimal power assignment and code hopping. Our framework is able to carry variable QoS multimedia traffic, efficiently serving a wide range of classes with CBR, VBR, ABR and UBR characteristics. Dynamic resource scheduling improves capacity and throughput performance. Other advantages are power saving and faster convergence to the desired power level.