skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Strategies for job allocation in a hypercube concurrent computer

Abstract

The author studies the problem of how to effectively use hypercube resources (processor) for the hypercube systems which supports multiple users. When each job arrives at hypercube, the operating system will allocate a dedicate free subcube to it. One way to attack this problem is to try to minimize finishing time of a sequence of jobs. The other way is to try to reduce the fragmentation, like memory fragmentation, occurs after subcubes allocation and release. Several strategies have been studied. These are non-preemptive scheduling, preemptive scheduling, and virtual subcube formation. The problem of scheduling k independent jobs on an n-dimensional hypercube system is to minimize finishing time, where each job J, is associates with a dimension d{sub i} and a processing time t{sub i}, meaning that J{sub i} requires a d{sub i}-dimensional subcube (call J{sub i} a d{sub i}-dimensional job) for t{sub i} units of time. This problem is NP - complete if no preemption is allowed. The author proposes a simple heuristic called LDLPT (Largest Dimension Largest Processing Time) for this problem and analyze its worse -case performance: the ratio of the heuristic finishing time to the optimal does not exceed 2-1/2{sup n-1}. If preemption is allowed, and O(k{supmore » 2}) algorithm is presented that decides if all k jobs can be finished by a given deadline. Using this algorithm, one may obtain a minimum-finishing-time schedule in polynomial time. The characteristic of the hypercube fragmentation is that even if a sufficient number of processors are available in the hypercube, they do not form a subcube large enough to accommodate an incoming job.« less

Authors:
Publication Date:
Research Org.:
Ohio State Univ., Columbus, OH (USA)
OSTI Identifier:
5923068
Resource Type:
Miscellaneous
Resource Relation:
Other Information: Thesis (Ph. D.)
Country of Publication:
United States
Language:
English
Subject:
99 GENERAL AND MISCELLANEOUS//MATHEMATICS, COMPUTING, AND INFORMATION SCIENCE; HYPERCUBE COMPUTERS; TASK SCHEDULING; ALGORITHMS; DATA TRANSMISSION; EXECUTIVE CODES; MEMORY DEVICES; MEMORY MANAGEMENT; COMMUNICATIONS; COMPUTER CODES; COMPUTERS; DATA PROCESSING; MATHEMATICAL LOGIC; PROCESSING 990200* -- Mathematics & Computers

Citation Formats

Chen, Guan-Ing. Strategies for job allocation in a hypercube concurrent computer. United States: N. p., 1989. Web.
Chen, Guan-Ing. Strategies for job allocation in a hypercube concurrent computer. United States.
Chen, Guan-Ing. 1989. "Strategies for job allocation in a hypercube concurrent computer". United States. doi:.
@article{osti_5923068,
title = {Strategies for job allocation in a hypercube concurrent computer},
author = {Chen, Guan-Ing.},
abstractNote = {The author studies the problem of how to effectively use hypercube resources (processor) for the hypercube systems which supports multiple users. When each job arrives at hypercube, the operating system will allocate a dedicate free subcube to it. One way to attack this problem is to try to minimize finishing time of a sequence of jobs. The other way is to try to reduce the fragmentation, like memory fragmentation, occurs after subcubes allocation and release. Several strategies have been studied. These are non-preemptive scheduling, preemptive scheduling, and virtual subcube formation. The problem of scheduling k independent jobs on an n-dimensional hypercube system is to minimize finishing time, where each job J, is associates with a dimension d{sub i} and a processing time t{sub i}, meaning that J{sub i} requires a d{sub i}-dimensional subcube (call J{sub i} a d{sub i}-dimensional job) for t{sub i} units of time. This problem is NP - complete if no preemption is allowed. The author proposes a simple heuristic called LDLPT (Largest Dimension Largest Processing Time) for this problem and analyze its worse -case performance: the ratio of the heuristic finishing time to the optimal does not exceed 2-1/2{sup n-1}. If preemption is allowed, and O(k{sup 2}) algorithm is presented that decides if all k jobs can be finished by a given deadline. Using this algorithm, one may obtain a minimum-finishing-time schedule in polynomial time. The characteristic of the hypercube fragmentation is that even if a sufficient number of processors are available in the hypercube, they do not form a subcube large enough to accommodate an incoming job.},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = 1989,
month = 1
}

Miscellaneous:
Other availability
Please see Document Availability for additional information on obtaining the full-text document. Library patrons may search WorldCat to identify libraries that may hold this item.

Save / Share:
  • Queueing system models of certain resource allocation policies in Distributed Computer Networks are analyzed in terms of an architecture comprised of five layers; user, processing, switching, multiplexing and communication. In the multiplex layer Integrated Voice/Data Multiplexing (IVDM) is analyzed. Voice talkspurts are serviced by either discarding, freeze-outs or queueing during periods of overload. Voice has priority over data. Data sources may be either local (high-speed access) or remote (low speed access). A two-dimensional matrix geometric solution determines data delay performance. Data delay is shown to be inversely proportional to the ratio of voice talkspurt to data packet duration. Remote datamore » has greater data delay while local data has greater delay variation. Additionally local data has greater peak queueing power while the peak queueing power fractional operating efficiency is greater for remote data. The percentage savings of integration over separate systems designed for equivalent performance is greatest when data is a small portion of the total system load, usually exceeding implementation overhead. In the switching layer the Fast Packet Switching (FPS) techniques of input queueing, input smoothing and output queueing are analyzed using a uniform queueing model. A new, more general switch architecture termed input queueing/input smoothing is defined, analyzed and shown to have performance that approaches that of a broadcast switch under reasonable design limits. A feasible implementation of this architecture by a hybrid optical/electronic design is described. In the processing layer static and dynamic computer resource allocation schemes are analyzed using queueing system models. Finally, divisible tasks tend to have lower delays but more variability than non-divisible ones. The percentage savings bound the amount of overhead which can be employed to allocate and divide tasks.« less
  • The subject of this study is formulation and performance evaluation of heuristic tack allocation algorithms for distributed computer systems. Both the formulation framework and evaluation environment are based on automata theory paradigms. The synthesis of task allocation algorithms and simulated distributed computations is based on the concept of a learning automation operating in a random environment. The evaluation environment is a simulator - interpreter based on formal specification models. The specification models are formal languages used to describe distributed computations and bus oriented hardware structures. Proposed learning heuristic task allocation algorithms perform global, dynamic, physically distributed, adaptive cooperative scheduling inmore » general purpose distributed computations executing on a hardware consisting of a number of processors connected to a shared bus. A number of simulation studies analyzed the performance of these algorithms under different modes of learning, in local and global cooperative environments and in stationary and nonstationary distributed computations. Furthermore, stability of these algorithms was analyzed under different operating conditions. Results of the study suggest that proposes algorithms may be a simple and cost - effective tool in performance optimization of target distributed systems.« less
  • The purpose of this grant was to purchase a hypercube computer to support DoD-sponsored research, and other research, in parallel numerical computation. The plans that were outlined in the proposal were followed very closely. A 32-node Intel iPSC-D5 hypercube computer with the funds from this grant, and have used the matching University contribution to support the maintenance of this computer. This computer has been used, over the past year, for research in parallel numerical computation in areas including global optimization, VLSI design, large sparse systems of linear equations, and eigenvalue problems, as well as for research in the languages formore » parallel computation and in the debugging and measurement of parallel programs.« less
  • A system of image processing and analysis utilities for a general-purpose hypercube topology concurrent multiprocessor system is described. The purpose of the system is to provide an efficient development and run-time environment for theoretical and experimental inquiry into computational vision for mobile and manipulative robots. An additional objective is to provide an environment with rich support for the development of concurrent computer vision algorithms. A number of principles for programming concurrent systems are embodied in the implementation of this system. In particular, all significant computation is performed by the hypercube proper, reserving the host processor exclusively for input/output functions. Onemore » consequence of this principle is that applications are developed by viewing the concurrent multiprocessor as a single computer with a special internal structure, rather than as a number of independent machines. Applications derive from a single source, rather than many, reducing development time without sacrificing run-time efficiency. This document serves as an introduction to the image processing and analysis system, its principles of operation, and as a reference manual for its use. 21 refs., 3 figs.« less
  • We have developed a new algorithm for implementing plasma particle-in-cell (PIC) simulation codes on concurrent processors with distributed memory. This algorithm, named the general concurrent PIC algorithm (GCPIC), has been used to implement an electrostatic PIC code on the 33-node JPL Mark III Hypercube parallel computer. To decompose at PIC code using the GCPIC algorithm, the physical domain of the particle simulation is divided into sub-domains, equal in number to the number of processors, such that all sub-domains have roughly equal numbers of particles. For problems with non-uniform particle densities, these sub-domains will be of unequal physical size. Each processormore » is assigned a sub-domain and is responsible for updating the particles in its sub-domain. This algorithm has led to a a very efficient parallel implementation of a well-benchmarked 1-dimensional PIC code. The dominant portion of the code, updating the particle positions and velocities, is nearly 100% efficient when the number of particles is increased linearly with the number of hypercube processors used so that the number of particles per processor is constant. For example, the increase in time spent updating particles in going from a problem with 11,264 particles run on 1 processor to 360,448 particles on 32 processors was only 3% (parallel efficiency of 97%). Although implemented on a hypercube concurrent computer, this algorithm should also be efficient for PIC codes on other parallel architectures and for large PIC codes on sequential computers where part of the data must reside on external disks. {copyright} 1989 Academic Press, Inc.« less