the proceedings contain 20 papers. the topics discussed include: DRAS: fast dynamic rescheduling scheme eliminating redundant computation;comparison of table join execution time for parallel DBMS and MapReduce;GPU acc...
the proceedings contain 20 papers. the topics discussed include: DRAS: fast dynamic rescheduling scheme eliminating redundant computation;comparison of table join execution time for parallel DBMS and MapReduce;GPU accelerated vessel segmentation using Laplacian eigenmaps;generating realistic network graph models for fault-tolerant algorithm evaluation;seamless multicore parallelism in MATLAB;a hierarchical library for user-defined schedulers;a computational performance investigation of Java concurrency using multi-threading on multi-cored processors;SC-FDE transmission with strong Doppler effects;on crosstalk-free BPC permutations routing in an optical variable-stage shuffle-exchange network;interpolation-based object decomposition and parallel computation method for large-scale computer-generated hologram;power-efficient load distribution in heterogeneous computing environments;and SalbNet: a self-adapting load balancing network.
Server Load Balancing (SLB) is a popular technique to build high-availability web services as offered from Google and Amazon for example. Credit based load balancing strategies have been proposed in the literature whe...
详细信息
Server Load Balancing (SLB) is a popular technique to build high-availability web services as offered from Google and Amazon for example. Credit based load balancing strategies have been proposed in the literature where the back end servers dynamically report a metric called Credit to the Load Balancer (LB) which reflects their current capacity. this enables the LB to adapt the load balancing strategy. the benefit of Credit based SLB has been shown by simulations, but up to now, it is not used in productive systems, since efficient implementations were missing. this paper presents the evaluation of an implementation of Credit based SLB, the so-called Self-Adapting Load Balancing Network (salbnet). We evaluate salbnet for a cluster of web servers. the measurements are done with a representative workload based on a Wikipedia trace and confirm the benefit of the self-adapting load balancing approach.
Analysis of existing research work indicates that preference for implementation of queries to structured data is given to parallel DBMS. MapReduce (MR) is perceived as supplementary to DBMS technology. We attempt to f...
详细信息
Analysis of existing research work indicates that preference for implementation of queries to structured data is given to parallel DBMS. MapReduce (MR) is perceived as supplementary to DBMS technology. We attempt to figure out behavior pattern of parallel rowstorage DBMS and MR system Hadoop on the example of Join task depending on the variation of the parameters that in other authors' experiments do not vary or differ from ours. this article presents detailed process models for table joins in the parallel row-storage DBMS and MRsystem, as well as the results of detailed calculation experiments performed on these models. the models were set up for various scalability schemes for MR (number of nodes) and DMBS (data volume in a node) and fragmentation of the joined tables by the primary key. the following parameters were varied: queried data selectivity, number of sorted resulting records and cardinality of the grouping attribute. the modeling results showed that withthe increase of the stored data volume parallel DBMS starts losing against MR-system at certain thresholds.
Computer Generated Holography (CGH) is considered a promising candidate for realizing 3D display with complete depth features. However, the realization of a largescale CGH requires a huge computation time. this paper ...
详细信息
Computer Generated Holography (CGH) is considered a promising candidate for realizing 3D display with complete depth features. However, the realization of a largescale CGH requires a huge computation time. this paper proposes a method for decomposing an input object into sub-objects and generating sub-holograms from them utilizing interpolation. the major advantage of this method is that the generation processes become mutually independent and can be executed in parallel without communication and synchronization. After presenting the theory of interpolation method, we will show how we can implement the method on GPU using data-parallel operations. We will also show the simulation and optical reconstruction results that verify the correctness of the method. Finally, we will present the preliminary results of our experiment by using GPU.
High performance servers of heterogeneous computing environments, as can be found in data centers for cloud computing, consume immense amounts of energy even though they are usually underutilized. In times when not al...
详细信息
High performance servers of heterogeneous computing environments, as can be found in data centers for cloud computing, consume immense amounts of energy even though they are usually underutilized. In times when not all computing capabilities are needed the task to be solved is how to distribute the computational load in a power-efficient manner. the question to be answered is, what load partitions should be assigned to each physical server so that all work is done with minimal energy consumption. this problem is closely related to the selection of physical servers that can be switched off completely to further reduce the power consumption. In this work, we present algorithms which calculate a power-efficient distribution of a divisible workload among multiple, heterogeneous physical servers. We assume a fully divisible load to calculate an optimized utilization of each server. Based on this distribution, an iterative process is carried out to identify servers, which can be switched off in order to further reduce the power consumption. Withthat information, workload (re)distribution can take place to partition appropriate subloads to the remaining servers. As before, the calculated partitioning minimizes the power consumption.
Task scheduling is very important for efficient execution of large-scale workflows. Static scheduling schemes achieve high performance when executing workflows in stable environments. However, the scheduling costs are...
详细信息
Task scheduling is very important for efficient execution of large-scale workflows. Static scheduling schemes achieve high performance when executing workflows in stable environments. However, the scheduling costs are very high for large-scale workflows and they may perform poorly if the system performance is changed dynamically. Dynamic rescheduling schemes support the dynamic performance changes because tasks are rescheduled using algorithms from static scheduling schemes when the performance is changed. thus, better performance can be achieved in workflow applications, although the scheduling costs will increase if the system performance is frequently changed. therefore, we propose a dynamic recursive adaptive scheduling scheme (DRAS) which can reduces the computational cost. DRAS uses RAS, which is a static scheme with low cost, as rescheduling algorithm. However, if DRAS uses RAS algorithm with no change, the cost increases since RAS is called many times. thus, we improved the RAS to eliminate the redundant computation. Furthermore, we modified RAS to improve the makespan when used for the rescheduling since DRAS may perform poorly in that case. the evaluation using an abstract simulation shows the makespan of DRAS decreases by 30%compared with a dynamic rescheduling scheme which uses the conventional RAS algorithm. Moreover, DRAS considerably reduces the scheduling time, achieving approximately 6 times speedup for workflows consisting of 1,000 tasks when the dynamic performance is frequently changed.
In order to improve the deficiency generated from uneven distribution of anchors in the distributed semidefinite programming (SDP) method, improved distributed method is proposed for solving Euclidean metric localizat...
详细信息
ISBN:
(纸本)9781479950799
In order to improve the deficiency generated from uneven distribution of anchors in the distributed semidefinite programming (SDP) method, improved distributed method is proposed for solving Euclidean metric localization problems that arise from large-scale wireless sensor networks (WSN). By introducing the change of factorization, nonlinear programming (NLP) model is presented on each subarea, and feasible direction algorithm is introduced for solving NLP problems, which can be executed in parallel. Numerical results on large-scale sensor network problems with more than 10000 nodes demonstrate that, the proposed method performs better than the distributed SDP method.
BLAST([1]) (Basic Local Alignment Search Tool) is a suite of programs used to identify similarity between genetic sequences. It is one of the most widely used tools in Bioinformatics. In recent years, withthe size of...
详细信息
ISBN:
(纸本)9781479950799
BLAST([1]) (Basic Local Alignment Search Tool) is a suite of programs used to identify similarity between genetic sequences. It is one of the most widely used tools in Bioinformatics. In recent years, withthe size of gene and protein sequence database increasing exponentially, BLAST has become both a data-intensive and a computation-intensive application. How to run BLAST rapidly with low cost has always been the hotspot to researchers. parallelization is one of the most important ways to resolve this problem. In this paper, a new approach for parallelizing BLAST based on a parallel processing framework called Robinia is presented. Compared withparallel version of BLAST presented before, Robinia-based BLAST has easy public accessibility and good scalability. Most importantly, it can support operation on WAN, this make it possible to integrate computation and storage resources on Internet to service for super-large scale BLAST projects. We implemented the Robinia-based BLAST and experimented on it using two different datasets. the results show that parallel BLAST based on Robinia can achieve linear speedup based on number of used nodes with good scalability and low cost.
Many industries nowadays use management and decision making based on artificial neural networks. However, the major drawback of neural networks lies in their time and computational complexity. the problem with computa...
详细信息
Many industries nowadays use management and decision making based on artificial neural networks. However, the major drawback of neural networks lies in their time and computational complexity. the problem with computational complexity could be eliminated using sharing of the computing needs on multiple computing nodes. this article focuses on the architectural design of a distributed system, which aims to solve large neural networks. the article describes the technology GPGPU and the next part of the article deals with an overview of methods for speeding up the calculation and distribution of artificial neural network. the main section describes the design of a model architecture description of the algorithm that allows correct data distribution on computational nodes.
BLAST[1] (Basic Local Alignment Search Tool) is a suite of programs used to identify similarity between genetic sequences. It is one of the most widely used tools in Bioinformatics. In recent years, withthe size of g...
详细信息
BLAST[1] (Basic Local Alignment Search Tool) is a suite of programs used to identify similarity between genetic sequences. It is one of the most widely used tools in Bioinformatics. In recent years, withthe size of gene and protein sequence database increasing exponentially, BLAST has become both a data-intensive and a computation-intensive application. How to run BLAST rapidly with low cost has always been the hotspot to researchers. parallelization is one of the most important ways to resolve this problem. In this paper, a new approach for parallelizing BLAST based on a parallel processing framework called Robinia is presented. Compared withparallel version of BLAST presented before, Robinia-based BLAST has easy public accessibility and good scalability. Most importantly, it can support operation on WAN, this make it possible to integrate computation and storage resources on Internet to service for super-large scale BLAST projects. We implemented the Robinia-based BLAST and experimented on it using two different datasets. the results show that parallel BLAST based on Robinia can achieve linear speedup based on number of used nodes with good scalability and low cost.
暂无评论