Hidehiro Kanemitsu

Clustering-Based Task Scheduling in a Large Number of Heterogeneous Processors

Parallelization paradigms for effective execution in a Directed Acyclic Graph (DAG) application have been widely studied in the area of task scheduling. Schedule length can be varied depending on task assignment policies, scheduling policies, and heterogeneity in terms of each processor and each communication bandwidth in a heterogeneous system. One disadvantage of existing task scheduling algorithms is that the schedule length cannot be reduced for a data intensive application. In this paper, we propose a clustering-based task scheduling algorithm called Clustering for Minimizing the Worst Schedule Length (CMWSL) to minimize the schedule length in a large number of heterogeneous processors. First, the proposed method derives the lower bound of the total execution time for each processor by taking both the system and application characteristics into account. As a result, the number of processors used for actual execution is regulated to minimize the Worst Schedule Length (WSL). Then, the actual task assignment and task clustering are performed to minimize the schedule length until the total execution time in a task cluster exceeds the lower bound. Experimental results indicate that CMWSL outperforms both existing list-based and clustering-based task scheduling algorithms in terms of the schedule length and efficiency, especially in data-intensive applications.

A Task Scheduling Method after Clustering for Data Intensive Jobs in Heterogeneous Distributed Systems

Several task clustering heuristics are proposed for allocating tasks in heterogeneous systems to achieve a good response time in data intensive jobs. However, one of the challenging problems is the process in task scheduling after task allocation by task clustering. We propose a task scheduling method after task clustering, leveraging worst schedule length (WSL) as an upper bound of the schedule length. In our proposed method, a task in a WSL sequence is scheduled preferentially to make the WSL smaller. Experimental results by simulation show that the response time is improved in several task clustering heuristics. In particular, our proposed scheduling method with the task clustering outperforms conventional list-based task scheduling methods.

Static Task Cluster Size Determination in Homogeneous Distributed Systems

In a distributed system, which consists of an unknown number of processors, it is important to derive an appropriate number of processors by which the good schedule length is obtained by a task scheduling. Many task clustering heuristics have been proposed to determine the number of processors and to minimize the schedule length for scheduling a directed acyclic graph (DAG) application. However, those heuristics are not aware of the actual number of existing processors. As a result, the number of processors determined by an existing task clustering may exceed that of actually existing processors. Therefore, conventional approaches adopt merging of each cluster for reducing the number of clusters at the expense of decreasing degree of task parallelism. In this paper, we present a static cluster size determination method, which derives the lower bound of the cluster size with considering the DAG structure and the task size to data size ratio to suppress the schedule length with the small number of processors. Our experimental evaluations by simulations show that the lower bound of each cluster size determined by the proposed method has a good impact on both the schedule length and the processor utilization.

A Monitoring Framework of WS-Resource Properties in WS-GRAM

Grid computing has been widely spread in pragmatic areas and scientific research areas in order to achieve more powerful computation than ordinary computation by single computer. Globus Toolkit4 (GT4), which has been the de facto standard middleware in Grid, gives to us many chances of submitting jobs, especially using WSRF technologies. However, in case of cluster type grid using GT4, rules for job execution depend on a local scheduler (LS) which interacts with WS-GRAM in GT4, so that clients can not reflect their own rules to behaviors in each worker node. In this paper, we formulate those potential problems and solutions based on our proposed architecture in, whose main feature is to monitor resource property (RP) in WS-GRAM and notify its status to worker nodes. Then we describe how to implement a mechanism for monitoring RP in WS-GRAM to achieve the transparent delivery of clients rules to worker nodes. Our experimental results show that our proposed architecture can be incorporated in WS-GRAM without difficulty.

On the method for realizing globally and locally accessible resource management for WS-GRAM

In recent years, Web service technologies are becoming more and more important communication scheme for grid environment. WSRF-based resource management components, such as WS-GRAM in GT4, lay emphasis on how to make existing procedures such as local scheduler be invoked via HTTP, so that they inevitably have two problems described below. (1) If assumed environment is cluster type grid, each worker node does not have access scheme to any information included in resource property of master node. (2) As a result, the contents of information sent to clients is forced by the local scheduler and therefore clients can not control job status in accordance with their own policy. If those two problems are resolved, any clients own resource management policy can be reflected on worker nodes. In this paper, we propose a method for resolving those problems in order to realize the interactive communication between clients and worker nodes for WSRF-based cluster grid.

Evaluation of task clustering algorithm by FFT for heterogeneous distributed system

In this paper, the evaluation result of a task clustering heuristic algorithm proposed for heterogeneous distributed systems is shown. The proposed heuristic algorithm is based on our original concept, known as “Worst Schedule Length (WSL)”. It derives the lower bound of the total execution time of the cluster for each processor using WSL, then the processor which contributes to minimize WSL is chosen as an assignment target. Task clustering is then performed to get minimal response time (i.e., minimal schedule length). We show that our proposed method has advantages over existing conventional approaches through the evaluation results.

Prior node selection for scheduling workflows in a heterogeneous system

Abstract Many workflow scheduling algorithms for heterogeneous systems have been developed to satisfy multiple requirements such as minimizing schedule length while maximizing throughput. In particular, in list-based scheduling approaches, the schedule length depends on the given nodes as well as the task allocation and ordering policies. This is because the scheduling priority is derived by averaging the execution time and communication time of the given nodes. If the set of nodes can be adjusted before the scheduling tasks, a small schedule length can be achieved. In this paper, we propose a prior node selection algorithm, called lower bound based candidate node selection (LBCNS) to select a subset of given nodes to minimize the schedule length while fairly scheduling each job. Our proposal has two approaches: (i) LBCNS_DEFAULT, which considers the job characteristics and each node’s performance, and (ii) priority-based LBCNS, which additionally takes each scheduling priority into account for a dedicated task scheduling algorithm. The experimental results of extensive simulations show that LBCNS_DEFAULT has the best fairness for scheduling multiple workflow jobs, while priority-based LBCNS achieves the minimum schedule length with the highest efficiency for a single workflow job and multiple workflow jobs.

An efficient file transfer using helper in bittorrent

Data transfer of large files over the Internet is often performed using Peer-to-Peer (P2P) technology. BitTorrent is the most popular P2P content distribution application. In general, BitTorrent users have much higher download bandwidth than upload bandwidth. It leads to bottleneck problem under the lack of the upload bandwidth, and it drastically degrades the performance of the overall system. In this paper, we propose a method to utilize idle upload bandwidth of helper in order to improve the file sharing efficiency on the BitTorrent. We also evaluate the efficiency of the proposed method by using simulation.

Effective use of computational resources in multicore distributed systems

In the last decades, many kinds of task execution models such as grid and cloud computing have been developed. In such distributed systems, each task is processed by respective processor in multicored computers e.g., household PCs which we can easily harness in recent years. If there is one policy to automatically decide the “best” combination and the number of processors (and computers), we effectively utilize those computational resources, thereby large number of jobs can be executed in parallel. In this paper, we propose a method for mapping of execution units for such environments. The method adopts a remapping technology after processor-execution unit mapping[6] is finished. Experimental comparisons by a simulation show the advantages of the proposed method.

A user-adaptive symbol presentation for non-verbal communication

This paper presents the symbol presentation method for Universal Design (UD). Current symbol presentations for UD focus on how to design a simple and perceptible symbol for as many users as possible. As a result, the same symbol is presented to every user and therefore some users with specific attributes (e.g., color-blindness, cultural or physical conditions) can not perceive the symbol appropriately. Thus, we propose the user-adaptive symbol presentation in which optimal components (color, shape and background) for a symbol is automatically selected using Analytic Hierarchy Process (AHP). We also present how compatibility among those components should be taken into account in order to be intuitively perceivable as a whole pictogram. The experimental comparisons show that our proposed user-adaptive symbol presentation has considerable improved satisfaction related to visual effects and distinction of color/shape than the conventional approach.