博弈树并行搜索算法

本文分析了博弈树并行搜索算法中的主变量分裂算法的同步开锁问题。     

异构集群环境下负载平衡调度算法研究与实现

围绕平衡负载这一目标,针对进程级并行任务的动态调度问题进行了研究,提出了一个异构集群环境下动态负载平衡算法,它结合了自适应数据采集与交换算法,有效的解决了服务器之间负载不平衡的问题,提高了系统的吞吐率。     

Web服务器集群的负载均衡中遗传算子的设计

应用遗传算法进行作业调度已被越来越多的学者关注。在Web服务器集群环境中,对于客户端的Web请求分配问题,采用常规的遗传算法进行负载均衡并不总是有效的,好的遗传算子对算法收敛性及收敛到最优解非常重要。基于集群环境中Web请求分配的特点,设计了有针对性的遗传算子,即改进的内外结合交叉算子和主动变异算子。模拟实验结果与分析表明这些算子对Web集群的请求分配是有效的。     

具有数目约束的负载均衡问题

考虑了具有数目约束的负载平衡问题的一种特殊情形,称之为2-半匹配问题。分析了此问题在3种目标函数下的计算复杂性,并设计了相应的近似算法。     

基于集群的负载平衡调度算法研究与实现

在集群系统的负载调度研究中,针对请求的服务时间变化大的特点,该文提出了一个动态反馈负载平衡算法,它结合内核中的加权连接调度算法,根据动态反馈回来的负载信息来调整服务器的权值,从而有效地解决服务器问的负载不平衡问题,提高了系统的吞吐率。     

SUPPLE: An efficient run-time support for non-uniform parallel loops

This paper presents SUPPLE (SUPort for Parallel Loop Execution), an innovative run-time support for the execution of parallel loops with regular stencil data references and non-uniform iteration costs. SUPPLE relies upon a static block data distribution to exploit locality, and combines static and dynamic policies for scheduling non-uniform iterations. It adopts, as far as possible, a static scheduling policy derived from the owner computes rule, and moves data and iterations among processors only if a load imbalance actually occurs. SUPPLE always tries to overlap communications with useful computations by reordering loop iterations and prefetching remote ones in the case of workload imbalance. The SUPPLE approach has been validated by many experimental results obtained by running a multi-dimensional flame simulation kernel on a 64-node Cray T3D. We have fed the benchmark code with several synthetic input data sets built on the basis of a load imbalance model. We have compared our results with those obtained with a CRAFT Fortran implementation of the benchmark.     

On weighted balls-into-bins games

We consider the well-known problem of randomly allocating m$m$ balls into n$n$ bins. We investigate various properties of single-choice games as well as multiple-choice games in the context of weighted balls. We are particularly interested in questions that are concerned with the distribution of ball weights, and the order in which balls are allocated. Do any of these parameters influence the maximum expected load of any bin, and if yes, then how?     

Optimal On-Line Algorithms to Minimize Makespan on Two Machines with Resource Augmentation

We study the problem of on-line scheduling on two uniformly related machines where the on-line algorithm has resources different from those of the off-line algorithm. We consider three versions of this problem, preemptive semi-online, non-preemptive on-line and preemptive on-line scheduling. For all these cases we design algorithms with best possible competitive ratios as functions of the machine speeds. This work was submitted as a part of the M.Sc. thesis of the second author. A preliminary version of this paper appeared in the proceedings of The First Workshop on Approximation and Online Algorithms (WAOA’03), pages 109–122.     

Deterministic parallel backtrack search

The backtrack search problem involves visiting all the nodes of an arbitrary binary tree given a pointer to its root subject to the constraint that the children of a node are revealed only after their parent is visited. We present a fast, deterministic backtrack search algorithm for a p-processor COMMON CRCW-PRAM, which visits any n-node tree of height h in time O((n/p+h)(logloglogp)²). This upper bound compares favourably with a natural Ω(n/p+h) lower bound for this problem. Our approach embodies novel, efficient techniques for dynamically assigning tree-nodes to processors to ensure that the work is shared equitably among them.     

Offline file assignments for online load balancing

We study a novel load balancing problem that arises in web search engines. The problem is a combination of an offline assignment problem, where files need to be (copied and) assigned to machines, and an online load balancing problem, where requests ask for specific files and need to be assigned to a corresponding machine, whose load is increased by this.We present simple deterministic algorithms for this problem and exhibit an interesting trade-off between the available space to make file copies and the obtainable makespan. We also give non-trivial lower bounds for a large class of deterministic algorithms and present a randomized algorithm that beats these bounds with high probability.     

Competitive distributed decision-making

We study several natural problems in distributed decision-making from the standpoint of competitive analysis; in these problems incomplete information is a result of the distributed nature of the problem, as opposed to the on-line mode of decision making that was heretofore prevalent in this area. In several simple situations of distributed scheduling, the competitive ratio can be computed exactly, and the different ratios can be used as a measure of the value of information and communication between decision-makers. In a more general distributed scheduling situation, we give tight upper and lower bounds on the competitive ratio achievable in the deterministic case, and give an optimal randomized algorithm with a much better competitive ratio.The research of Xiaotie Deng was supported by an NSERC grant and that of C. H. Papadimitriou was supported by an NSF grant.     

THE LOAD DISTRIBUTION PROBLEM IN A PROCESSOR RING

Abstract

Given a global picture of the system load and the average load, the load distribution problem is to find a suitable schedule, consisting of the amount of excess load to transfer along every edge, so that the system load can be balanced in minimal time by executing the schedule. We study this problem for the ring topology We discuss some existing algorithms, show how they fall short of being able to generate optimal schedules, and present a simple algorithm that would generate an optimal schedule for any given system load instance. This simple algorithm relies on an existing algorithm to create a search window in which the optimal solution is to be found.     

Load balancing algorithms based on gradient methods and their analysis through algebraic graph theory

The main results of this paper are based on the idea that most load balancing algorithms can be described in the framework of optimization theory. It enables to involve classical results linked with convergence, its speed and other elements. We emphasize that these classical results have been found independently and till now this connection has not been shown clearly. In this paper, we analyze the load balancing algorithm based on the steepest descent algorithm. The analysis shows that the speed of convergence is determined by eigenvalues of the Laplacian for the graph of a given load balancing system. This consideration also leads to the problems of choosing an optimal structure for a load balancing system. We prove that these optimal graphs have special Laplacians: the multiplicities of their minimal and maximal positive eigenvalues must be greater than one. Such a property is essential for strongly regular graphs, investigated in algebraic graph theory.     

Improved Bounds for On-Line Load Balancing

We consider the following load balancing problem. Jobs arrive on-line and must be assigned to one of m machines thereby increasing the load on that machine by a certain weight. Jobs also depart on-line. The goal is to minimize the maximum load on any machine, the load being defined as the sum of the weights of the jobs assigned to the machine divided by the machine capacity. The scheduler also has the option of preempting a job and reassigning it to another machine. Whenever a job is assigned or reassigned to a machine, the on-line algorithm incurs a reassignment cost depending on the job. For arbitrary reassignment costs and identical machines, we present an on-line algorithm with a competitive ratio of 3.5981 against current load , i.e., the maximum load at any time is less than 3.5981 times the lowest achievable load at that time. Our algorithm also incurs a reassignment cost less than 6.8285 times the cost of assigning all the jobs. For arbitrary reassignment costs and related machines we present an algorithm with a competitive ratio of 32 and a reassignment factor of 79.4. We also describe algorithms with better performance guarantees for some special cases of the problem. Received August 24, 1996; revised August 6, 1997.     

Complexity analysis of a decentralised graph colouring algorithm

Colouring a graph with its chromatic number of colours is known to be NP-hard. Identifying an algorithm in which decisions are made locally with no information about the graph's global structure is particularly challenging. In this article we analyse the complexity of a decentralised colouring algorithm that has recently been proposed for channel selection in wireless computer networks.     

An efficient retrieval selection algorithm for video servers with random duplicated assignment storage technique

Random Duplicated Assignment (RDA) is an approach in which video data is stored by assigning a number of copies of each data block to different, randomly chosen disks. It has been shown that this approach results in smaller response times and lower disk and RAM costs compared to the well-known disk stripping techniques. Based on this storage approach, one has to determine, for each given batch of data blocks, from which disk each of the data blocks is to be retrieved. This is to be done in such a way that the maximum load of the disks is minimized. The problem is called the Retrieval Selection Problem (RSP). In this paper, we propose a new efficient algorithm for RSP. This algorithm is based on the breadth-first search approach and is able to guarantee optimal solutions for RSP in O(n²+mn), where m and n correspond to the number of data blocks and the number of disks, respectively. We will show that our proposed algorithm has a lower time complexity than an existing algorithm, called the MFS algorithm.     

Load balancing strategies for a parallel ray-tracing system based on constant subdivision

Static and dynamic load balancing strategies for a multiprocessor system for a ray tracing algorithm based on constant subdivision are presented. An object space is divided into regular cubes (subspaces), whose boundary planes are perpendicular to the coordinate axes, and these are allocated to the processors in the system. Here, load balancing among the processors is the most important problem. Firstly, in a category of static load balancing, strategies for mapping the subspaces into the processors are evaluated by simulation. Moreover, we propose a hierarchical multiprocessor system in order to realize dynamic load balancing with the static one. Its architecture can overcome the limitation of the static load balancing in a large scale multiprocessor system.     

Data migration on parallel disks: Algorithms and evaluation

Our work is motivated by the problem of managing data on storage devices, typically a set of disks. Such storage servers are used as web servers or multimedia servers, for handling high demand for data. As the system is running, to exhibit good performance, it needs to respond dynamically to changes in demand for different data items. There are known algorithms for mapping demand to a layout. When the demand changes, a new layout can be computed. In this work we study thedata migration problem, which arises when we need to change one layout to another quickly. This problem has been studied earlier where for each disk a new layout has been prescribed. However, to apply these algorithms effectively, we identify another problem that we refer to as the correspondence problem, whose solution has a significant impact on the overall solution for the data migration problem. We study algorithms for the data migration problem in more detail and identify variations of the basic algorithm that seem to improve performance in practice, even though some of the variations have poor worst-case behavior. This research was supported by the NSF Awards CCR-0113192 and EIA-0091474 as well as the Okawa Research Award. This work made use of Integrated Media Systems Center Shared Facilities supported by the National Science Foundation under Cooperative Agreement No. EEC-9529152; any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect those of the National Science Foundation. This work was done while Svetlana Shargorodskaya was at the University of Maryland.