Divisible Load Research Articles

Experimental evaluation of a multi-installment scheduling strategy based on divisible load paradigm for SAR image reconstruction on a distributed computing infrastructure

Radar loads, especially Synthetic Aperture Radar (SAR) image reconstruction loads use a large volume of data collected from satellites to create a high-resolution image of the earth. To design near-real-time applications that utilise SAR data, speeding up the image reconstruction algorithm is imperative. This can be achieved by deploying a set of distributed computing infrastructures connected through a network. Scheduling such complex and large divisible loads on a distributed platform can be designed using the Divisible Load Theory (DLT) framework. We performed distributed SAR image reconstruction experiments using the SLURM library on a cloud virtual machine network using two scheduling strategies, namely the Multi-Installment Scheduling with Result Retrieval (MIS-RR) strategy and the traditional EQual-partitioning Strategy (EQS). The DLT model proposed in the MIS-RR strategy is incorporated to make the load divisible. Based on the experimental results and performance analysis carried out using different pixel lengths, pulse set sizes, and the number of virtual machines, we observe that the time performance of MIS-RR is much superior to that of EQS. Hence the MIS-RR strategy is of practical significance in reducing the overall processing time, and cost, and in improving the utilisation of the compute infrastructure. Furthermore, we note that the DLT-based theoretical analysis of MIS-RR coincides well with the experimental data, demonstrating the relevance of DLT in the real world.

Extension of Divisible-Load Theory from Scheduling Fine-Grained to Coarse-Grained Divisible Workloads on Networked Computing Systems

The big data explosion has sparked a strong demand for high-performance data processing. Meanwhile, the rapid development of networked computing systems, coupled with the growth of Divisible-Load Theory (DLT) as an innovative technology with competent scheduling strategies, provides a practical way of conducting parallel processing with big data. Existing studies in the area of DLT usually consider the scheduling problem with regard to fine-grained divisible workloads. However, numerous big data loads nowadays can only be abstracted as coarse-grained workloads, such as large-scale image classification, context-dependent emotional analysis and so on. In view of this, this paper extends DLT from fine-grained to coarse-grained divisible loads by establishing a new multi-installment scheduling model. With this model, a subtle heuristic algorithm was proposed to find a feasible load partitioning scheme that minimizes the makespan of the entire workload. Simulation results show that the proposed algorithm is superior to the up-to-date multi-installment scheduling strategy in terms of achieving a shorter makespan of workloads when dealing with coarse-grained divisible loads.

Maximizing Heterogeneous Server Utilization with Limited Availability Times for Divisible Loads Scheduling on Networked Systems

Most of the available divisible-load scheduling models assume that all servers in networked systems are idle before workloads arrive and that they can remain available online during workload computation. In fact, this assumption is not always valid. Different servers on networked systems may have heterogenous available times. If we ignore the availability constraints when dividing and distributing workloads among servers, some servers may not be able to start processing their assigned load fractions or deliver them on time. In view of this, we propose a new multi-installment scheduling model based on server availability time constraints. To solve this problem, we design an efficient heuristic algorithm consisting of a repair strategy and a local search strategy, by which an optimal load partitioning scheme is derived. The repair strategy guarantees time constraints, while the local search strategy achieves optimality. We evaluate the performance via rigorous simulation experiments and our results show that the proposed algorithm is suitable for solving large-scale scheduling problems employing heterogeneous servers with arbitrary available times. The proposed algorithm is shown to be superior to the existing algorithm in terms of achieving a shorter makespan of workloads.

Optimum scheduling in fog computing using the Divisible Load Theory (DLT) with linear and nonlinear loads

Fog computing has been proposed to reduce the latency of cloud computing. One of the most common challenges in fog computing is scheduling. Suitable scheduling increases the efficiency of fog computing. There is a wide range of methods of scheduling that have been presented by the researchers each with its strengths and weaknesses. This paper presents a novel scheduling method for fog computing based on the Divisible Load Theory (DLT). The proposed method has significant benefits for loads that are arbitrarily divisible and for huge and intense data. First, the proposed method is modeled, and then, based on DLT, the related closed form is proposed for both linear and nonlinear modes. Subsequently, the closed forms are solved, and based on that, an innovative algorithm for the partitioning and distribution of fractions of an arbitrary divisible load among nodes in a fog environment is proposed. The performance of the proposed method is compared with existing algorithms. The simulated results show that using the DLT method significantly optimizes the performance a minimum of seven times. It reduces the finish time (about eight times in linear and eighty-five times in nonlinear loads) and increases the speedup (about seven times in linear and one hundred thirty times in nonlinear loads).

Optimized scheduling of multi‐user Map‐Reduce jobs in heterogeneous environment

SummaryMap‐Reduce is a programming paradigm widely used for data intensive applications in distributed computing. Divisible load theory (DLT) is another model to partition divisible load for parallel processing. In this work, a new job scheduler named virtual job scheduler (VJS) is developed to schedule the MapReduce jobs based on DLT. VJS constructs a virtual job set from the queue of jobs awaiting execution by considering the CPU and IO resource utilization levels of each job. The core of VJS is in its partitioning algorithm. Two novel partitioning algorithms, namely, two level successive partitioning (TLSP) and predictive partitioning (PRED) have been proposed. TLSP applies DLT to both map and reduce phase in succession. This second load partitioning performed during the comparatively longer reduce phase, exploits the advantage of DLT better. PRED is a modification of TLSP aimed at optimizing the overall schedule length, by taking the idle time of the reducers into consideration. Evaluation of these two models across various execution environments has indicated a profound decrease in the wait time of reducers and as a result has shown a significant reduction in the makespan of the whole job as such. VJS altered to incorporate PRED partitioning algorithm has proved to be suitable for heterogeneous environment, with high resource utilization compared with the default Hadoop MapReduce scheduler.

Optimum Scheduling in Fog Computing Using the Divisible Load Theory (Dlt) with Linear and Nonlinear Loads

Optimum Scheduling in Fog Computing Using the Divisible Load Theory (Dlt) with Linear and Nonlinear Loads

Load Balancing in Cloud Computing Empowered with Dynamic Divisible Load Scheduling Method

The need to process and dealing with a vast amount of data is increasing with the developing technology. One of the leading promising technology is Cloud Computing, enabling one to accomplish desired goals, leading to performance enhancement. Cloud Computing comes into play with the debate on the growing requirements of data capabilities and storage capacities. Not every organization has the financial resources, infrastructure & human capital, but Cloud Computing offers an affordable infrastructure based on availability, scalability, and cost-efficiency. The Cloud can provide services to clients on-demand, making it the most adapted system for virtual storage, but still, it has some issues not adequately addressed and resolved. One of those issues is that load balancing is a primary challenge, and it is required to balance the traffic on every peer adequately rather than overloading an individual node. This paper provides an intelligent workload management algorithm, which systematically balances traffic and homogeneously allocates the load on every node & prevents overloading, and increases the response time for maximum performance enhancement.

Integrating Amdahl-like Laws and Divisible Load Theory

A simple means of integrating the characteristics of networked processors under divisible loads into Amdahl’s Law is presented. Amdahl’s Law serves as an upper bound to these speedup results. Amdahl’s Law with divisible load processing characteristics included serves as an upper bound to speedup for any model taking into consideration more detailed peculiarities of real systems such as the overhead of task creation, synchronization, resource contention and memory issues.

A multi-criteria approach to time cheating in the divisible load scheduling

The divisible load scheduling (DLS) can be considered as a special class of scheduling model in the area of distributed and parallel systems. According to the DLS, the computations and communications can be divided into some arbitrarily independent fragments in which each fragment can be computed independently by a processor. The basic assumptions of the traditional DLS models are $$w_j\le w_{j+1}$$ and $$z_j\le z_{j+1}$$ for all $$j\le m$$ ( $$z_j$$ and $$w_j$$ are the rates of communication and computation of the jth processor, respectively). These presumptions are not acceptable if the processors do not report their real rates of communication or computation. The problem that a processor may not report its real rates of communication or computation is called cheating problem. This paper has a multi-criteria approach to the time cheating problem in the area of the DLS. The multi-criteria DLS model is a new paradigm in the area of DLS and its applications. The main contribution of this paper is to reduce the effects of time cheating in a DLS scheduling model.

Time–energy trade-offs in processing divisible loads on heterogeneous hierarchical memory systems

We analyze time and energy performance of distributed computations in heterogeneous systems with hierarchical memory. Different levels of memory hierarchy have different time and energy efficiency. Core memory may be too small to hold whole load to be processed, while computations using external storage are expensive in time and energy. In order to avoid the costs of processing the load in the external memory, it is allowed that the load is distributed to the worker processors in multiple installments. A minimum energy solution is found by use of mixed integer linear programming under a limit on schedule length. Two types of fast heuristics with several variants are also examined. The trade-off between the criteria of processing time and energy is studied. Key features of optimum solutions are analyzed. It is shown that holding machines in a diverse set of energy modes and limited use of the out-of-core memory can be beneficial for the time and energy performance. The proposed scheduling algorithms are evaluated in the terms of solution quality and runtimes.

Adapting Market-Oriented Policies for Scheduling Divisible Loads on Clouds

Cloud computing has become an important alternative for solving big data processing. Nowadays, cloud service providers usually offer users a virtual machine with various combinations of prices. As each user has different circumstances, the problem of choosing the cost-minimized combination under a deadline constraint as well as user's preference is becoming more complex. This article is concerned with the investigation of adapting a user's preference policies for scheduling real-time divisible loads in a cloud computing environment. The workload allocation approach used in this research is using Divisible Load Theory. The proposed algorithm aggregates resources into groups and optimally distributes the fractions of load to the available resources according to user's preference. The proposed algorithm was evaluated by simulation experiments and compared with the baseline approach. The result obtained from the proposed algorithm reveals that a significant reduction in computation cost can be attained when the user's preferences are low priority.

Scheduling divisible loads with time and cost constraints

In distributed computing, divisible load theory provides an important system model for allocation of data-intensive computations to processing units working in parallel. The main task is to define how a computation job should be split into parts, to which processors those parts should be allocated and in which sequence. The model is characterized by multiple parameters describing processor availability in time, transfer times of job parts to processors, their computation times and processor usage costs. The main criteria are usually the schedule length and cost minimization. In this paper, we provide the generalized formulation of the problem, combining key features of divisible load models studied in the literature, and prove its NP-hardness even for unrestricted processor availability windows. We formulate a linear program for the version of the problem with a fixed number of processors. For the case with an arbitrary number of processors, we close the gaps in the study of special cases, developing efficient algorithms for single criterion and bicriteria versions of the problem, when transfer times are negligible.

Efficient distributed average consensus in wireless sensor networks

Computing the distributed average consensus in Wireless Sensor Networks (WSNs) is investigated in this article. This problem, which is both natural and important, plays a significant role in various application fields such as mobile agents and fleet vehicle coordination, network synchronization, distributed voting and decision, load balancing of divisible loads in distributed computing network systems, and so on. By and large, the average consensus’ objective is to have all nodes in the network converged to the average value of the initial nodes’ measurements based only on local nodes’ information states. In this paper, we introduce a fully distributed algorithm to average the sensed data within the network itself. The network may be large since we never broadcast over all its nodes. Unlike earlier works, when a node detects a load (scalar value) imbalance in its closed neighborhoods during the average process, instead of sending parsimonious amount of load values from highly loaded nodes to less loaded ones, we move a large amount of load values by involving parallel atomic transactions between mutually exclusive pairs of neighbors. This improves the global convergence time speedup with low-cost communication and minimal energy consumption. First, we give the convergence proof of the distributed consensus process, and next we provide some experimental results based on NS3 framework to assess the behavior of the proposed algorithm.

Divisible load scheduling of image processing applications on the heterogeneous star and tree networks using a new genetic algorithm

SummaryThe divisible load scheduling of image processing applications on the heterogeneous star and multi‐level tree networks is addressed in this paper. In our platforms, processors and network links have different speeds. In addition, computation and communication overheads are considered. A new genetic algorithm for minimizing the processing time of low‐level image applications using divisible load theory is introduced. The closed‐form solution for the processing time, the image fractions that should be allocated to each processor, the optimum number of participating processors, and the optimal sequence for load distribution are derived. The new concept of equivalent processor in tree network is introduced and the effect of different image and kernel sizes on processing time and speed up are investigated. Finally, to indicate the efficiency of our algorithm, several numerical experiments are presented.

Shared processor scheduling of multiprocessor jobs

We study a problem of shared processor scheduling of multiprocessor weighted jobs. Each job can be executed on its private processor and simultaneously on possibly many processors shared by all jobs. This simultaneous execution reduces their completion times due to the processing time overlap. Each of the m shared processors may charge a different fee but otherwise the processors are identical. The goal is to maximize the total weighted overlap of all jobs. This is a key problem in subcontractor scheduling in extended enterprises and supply chains, and in divisible load scheduling in computing. We introduce synchronized schedules that complete each job that uses some shared processor at the same time on its private and on the shared processors. We prove that, quite surprisingly, the synchronized schedules include optimal ones. We obtain an α-approximation algorithm that runs in strongly polynomial time for the problem, where α=1/2+1/(4(m+1)). This improves the 1/2-approximation reported recently in the literature to 5/8-approximation for a single shared processor problem, m=1. The computational complexity of the problem, both in case of single and multi-shared processor, remains open. We show however an LP-based optimal algorithm for antithetical instances where for any pair of jobs j and i, if the processing time of j is smaller than or equal to the processing time of i, then the weight of j is greater than or equal to the weight of i.

Energy Harvesting Network With Wireless Distributed Computing

Bulky processing tasks are expected to burden the limited resources of energy harvesters by draining the stored energy, and thereby, reaching rapidly to energy causality constraint. In such scenario, energy harvesters flip into sleep mode, and thereby, the execution time of the next task will be delayed until the energy harvesters revert back into active mode. To tackle this problem, this paper proposes a novel energy harvesting network (EHN) that deploys wireless distributed computing (WDC) network within the decision making process (DMP). The DMP is formulated as constrained partially observable Markov decision process in order to enable the energy harvesters to act under uncertainty. Furthermore, various challenges of WDC networks, e.g., nominating the collaborating nodes and task allocation, have been addressed herein. Unlike conventional research works on WDC networks, a system model is proposed for WDC network based on divisible load theory instead of graph theory. In addition, an adaptive task allocation algorithm is proposed to distribute the task efficiently among the collaborating nodes. Finally, the novel EHN system is analyzed and compared against the conventional research works on WDC, offloading computing, and local computing-EHN, where the proposed system is found to outperform in terms of energy and delay.

Wireless Photoelectric Sensor Network Scheduling Algorithm Based on Divisible Load

To discuss the divisible load scheduling in wireless photoelectric sensor networks, a load scheduling algorithm called EDDLT based on residual energy landscape is proposed. In the algorithm, the constant of the time needed for the induction and reporting unit data is adjusted to the variable parameters based on the residual energy. In addition, the ratio of the initial energy to the residual energy is used to carry out the effective load scheduling. By using this algorithm, when the load is allocated to each sensor node, the remaining energy of nodes is considered, and a lighter load is allocated to sensor nodes with less residual energy. EDDLT, compared to the standard divisible load scheduling method SDLT, the number of execution rounds is greatly increased when the first sensor node is dead. The experimental results showed that EDDLT had a certain effect on prolonging the lifetime of wireless photoelectric sensor networks. To sum up, the scheduling algorithm has good performance in exploring divisible load.

COST-AWARE REAL-TIME DIVISIBLE LOADS SCHEDULING IN CLOUD COMPUTING

Cloud computing has become an important alternative for solving large scale resource- intensive problems in science, engineering, and analytics. Resource management play an important role in improving the quality of service (QoS). This paper is concerned with the investigation of scheduling strategies for divisible loads with deadlines constraints upon heterogeneous processors in a cloud computing environment. The workload allocation approach presents in this paper is using Divisible Load Theory (DLT). It is based on the fact that the computation can be partitioned into some arbitrary sizes and each partition can be processed independently of each other. Through series of simulation against the baseline strategies, it can be found that the worker selection order in the service pool and the amount of fraction load assigned to each of them have significant effects on the total computation cost.

Time Cheating in Divisible Load Scheduling: Sensitivity Analysis, Results and Open Problems

There is broad literature in relation to the Divisible Load Scheduling (DLS). The DLS is based on the fact that the computation and communication can be separated into some arbitrary independent parts, in which each part can be processed by an independent processor. The existing DLS models supposes that the processors report their true communication and computation rates to the root processor. It means the traditional DLS assumes that the processors do not cheat the root processor. In fact, in the real applications, the processors may cheat the algorithm it means the processors might not report their true computation or communication rates. However, the computation rate-cheating problem has been studied in several research. This paper reviews the research in relation to the communication and computation rate-cheating problems. The paper also investigates the sensitivity analysis of the DLS where the processors cheat the root processor for the first time.

Dynamic scheduling strategy with efficient node availability prediction for handling divisible loads in multi-cloud systems

With large resource capacity, clouds have become a primary infrastructure for users to store big amount of data and perform large-scale computations. With the increasing demands and diversity of applications’ requirements, users are facing a fundamental problem of management of resources reserved from clouds. Particularly, the problem of load scheduling is the most important since it directly affects the performance of the system. Designing an efficient scheduling strategy for minimizing the total processing time of loads is challenging since it has to consider many intrinsic characteristics of the system such as the availability and heterogeneity of computing nodes, network topology and capacity. In this paper, we propose a novel architecture of a multi-cloud system that can satisfy complex requirements of users’ applications. Based on this architecture, we propose a dynamic scheduling strategy (DSS) that integrates the Divisible Load Theory and node availability prediction techniques to achieve high performance. We conduct intensive simulations to evaluate the performance of the proposed scheduling strategy. The results show that the proposed scheduling strategy outperforms the baseline schemes by reducing the total processing time of loads up to 44.60%. The results also provide useful insights on the applicability of the proposed approach in realistic scenarios.