Strategies for dynamic load balancing on highly parallel computers

Willebeek-LeMair, M.; Reeves, Anthony P.

doi:10.1109/71.243526

Cited by 439 publications

(241 citation statements)

References 14 publications

Supporting

Mentioning

233

Contrasting

Unclassified

Order By: Relevance

“…All of these components have been studied extensively in the PDES literature, leading to the development of a large number of dynamic load-balancing algorithms. An exhaustive review of these load-balancing algorithms is beyond the scope of this article (but see, e.g., [121][122][123][124][125] for related work on this topic).…”

Section: Dynamic Process Migrationmentioning

confidence: 99%

“…According to Willebeek-Lemair and Reeves [121], dynamic load balancing can be organized as a procedure with four major components, including (1) processor load evaluation; (2) load balancing profitability determination; (3) load migration strategy, and (4) load selection strategy. All of these components have been studied extensively in the PDES literature, leading to the development of a large number of dynamic load-balancing algorithms.…”

Section: Dynamic Process Migrationmentioning

confidence: 99%

“…For example, DEVS/Grid [121] implements a grid-enabled DEVS simulator following a layered approach. Another grid example is described in vGrid [174], which divides the model into components that can be grouped together to form a virtual computational unit.…”

Section: Web-based Pdesmentioning

confidence: 99%

See 2 more Smart Citations

Synchronization methods in parallel and distributed discrete-event simulation

Jafer

Liu

Wainer

2013

Simulation Modelling Practice and Theory

View full text Add to dashboard Cite

a b s t r a c tThis work attempts to provide insight into the problem of executing discrete event simulation in a distributed fashion. The article serves as the state of the art in Parallel DiscreteEvent Simulation (PDES) by surveying existing algorithms and analyzing the merits and drawbacks of various techniques. We discuss the main characteristics of existing synchronization methods for parallel and distributed discrete event simulation. The two major categories of synchronization protocols, namely conservative and optimistic, are introduced and various approaches within each category are presented. We also present the latest efforts towards PDES on emerging platforms such as heterogeneous multicore processors, Web services, as well as Grid and Cloud environment.Ó 2012 Elsevier B.V. All rights reserved. Parallel discrete-event simulationWith the computing power and advanced software available today, modeling and simulation has become a cost-effective tool for detailed analysis of a broad array of natural and artificial systems. Parallel and distributed simulation is widely accepted as the technology of choice to speed up large-scale discrete-event simulation and to promote reusability and interoperability of simulation components. Parallel Discrete-Event Simulation (PDES) has received increasing interest as simulations become more time consuming and geographically distributed. A rich literature has already been developed in the last three decades, taking advantage of the increasing availability of highly parallel and distributed computing platforms. It has been several years since the last survey by Perumalla [5] and a refreshed review of the latest developments would be needed for us to ponder new research initiatives, especially on emerging platforms such as many-core processors, Internetscale simulation environments, and cloud-based virtualized infrastructures.In parallel and distributed simulations, the entire simulation task is divided into a set of smaller subtasks with each executed on a different processor or node. Hence, the simulation system is viewed as a collection of concurrent processes, each modeling a different part of the physical system and executing on a dedicated processor in a sequential fashion. These processes communicate with each other by exchanging time-stamped event messages. An event refers to an update to simulation system state at a specific simulation time instant. Throughout the simulation, events arrive at destination processes, and depending on the delivery ordering system of the simulation, they are processed differently. The two commonly used orderings are (1) receive-order and (2) timestamp-order. With the first type, events are delivered to the destination processes when they arrive at the destination. On the other hand, with the timestamp-order, events are delivered in non-decreasing order of their timestamp, requiring runtime checks and buffering to ensure such ordering.1569-190X/$ -see front matter Ó

show abstract

Section: Dynamic Process Migrationmentioning

confidence: 99%

Section: Dynamic Process Migrationmentioning

confidence: 99%

See 1 more Smart Citation

Synchronization methods in parallel and distributed discrete-event simulation

Jafer

Liu

Wainer

2013

Simulation Modelling Practice and Theory

View full text Add to dashboard Cite

show abstract

“…Our strategy is adapted from the local diffusive load balancing strategy introduced in [13]. It is a receiver-initiated strategy, which is found to be more efficient in [13].…”

Section: Load Balance Decision Strategymentioning

confidence: 99%

Efficient Dynamic Operator Placement in a Locally Distributed Continuous Query System

Zhou

Ooi

Tan

et al. 2006

On the Move to Meaningful Internet Systems 2006: CoopIS, DOA, GADA, and ODBASE

View full text Add to dashboard Cite

Abstract. In a distributed processing environment, the static placement of query operators may result in unsatisfactory system performance due to unpredictable factors such as changes of servers' load, data arrival rates, etc. The problem is exacerbated for continuous (and long running) monitoring queries over data streams as any suboptimal placement will affect the system for a very long time. In this paper, we formalize and analyze the operator placement problem in the context of a locally distributed continuous query system. We also propose a solution, that is asynchronous and local, to dynamically manage the load across the system nodes. Essentially, during runtime, we migrate query operators/fragments from overloaded nodes to lightly loaded ones to achieve better performance. Heuristics are also proposed to maintain good data flow locality. Results of a performance study shows the effectiveness of our technique.

show abstract

“…Assignment may be either deterministic, as with the dimension exchange/diffusion [6] and gradient methods, stochastic, or optimization based. A comparison of several deterministic methods is provided by Willeback-LeMain and Reeves [12].…”

Section: Introductionmentioning

confidence: 99%

Dynamic Time Delay Models for Load Balancing. Part I: Deterministic Models

Birdwell

Chiasson

Tang

et al. 2004

Lecture Notes in Computational Science and Engineering

View full text Add to dashboard Cite

Summary. Parallel computer architectures utilize a set of computational elements (CE) to achieve performance that is not attainable on a single processor, or CE, computer. A common architecture is the cluster of otherwise independent computers communicating through a shared network. To make use of parallel computing resources, problems must be broken down into smaller units that can be solved individually by each CE while exchanging information with CEs solving other problems.Effective utilization of a parallel computer architecture requires the computational load to be distributed more or less evenly over the available CEs. The qualifier "more or less" is used because the communications required to distribute the load consume both computational resources and network bandwidth. A point of diminishing returns exists.In this work, a nonlinear deterministic dynamic time-delay systems is developed to model load balancing in a cluster of computer nodes used for parallel computations. This model is then compared with an experimental implementation of the load balancing algorithm on a parallel computer network.

show abstract

Strategies for dynamic load balancing on highly parallel computers

Cited by 439 publications

References 14 publications

Synchronization methods in parallel and distributed discrete-event simulation

Synchronization methods in parallel and distributed discrete-event simulation

Efficient Dynamic Operator Placement in a Locally Distributed Continuous Query System

Dynamic Time Delay Models for Load Balancing. Part I: Deterministic Models

Contact Info

Product

Resources

About