Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Computational Load Balancing”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 73 records · Page 4

Load-balanced sparse MTTKRP on GPUs

Sparse matricized tensor times Khatri-Rao product (MT- TKRP) is one of the most computationally expensive kernels in sparse tensor computations. This work focuses on optimizing the MTTKRP for floating point operations, storage, and scalability. We begin by identifying the performance bottlenecks in directly extending the state-of-the-art CSF (compressed sparse fiber) formats from CPUs to GPUs. Our detailed analysis over the recently proposed formats shows that the lower bounds on storage and flop counts can vary significantly depending on the structure of the sparse tensor. To address this, we propose a load balanced, computation and storage-efficient scheme, HYB, which combines the best of COO (coordinate), CSF and CSL (compressed slice). With these enhancements, our GPU framework significantly out- performs the current formats on both CPU and GPU platforms.

Nisa, Israt↗

Technical note: Optimization for improved tube-loading efficiency in the dual-energy computed tomography coupled with balanced filter method

Purpose: This article describes the spectral optimization of dual-energy computed tomography using balanced filters (bf-DECT) to reduce the tube loadings and dose by dedicating to the acquisition of electron density information, which is essential for treatment planning in radiotherapy. Methods: For the spectral optimization of bf-DECT, the author calculated the beam-hardening error and air kerma required to achieve a desired noise level in an electron density image of a 50-cm-diameter cylindrical water phantom. The calculation enables the selection of beam parameters such as tube voltage, balanced filter material, and its thickness. Results: The optimal combination of tube voltages was 80 kV/140 kV in conjunction with Tb/Hf and Bi/Mo filter pairs; this combination agrees with that obtained in a previous study [M. Saito, ''Spectral optimization for measuring electron density by the dual-energy computed tomography coupled with balanced filter method,'' Med. Phys. 36, 3631-3642 (2009)], although the thicknesses of the filters that yielded a minimum tube output were slightly different from those obtained in the previous study. The resultant tube loading of a low-energy scan of the present bf-DECT significantly decreased from 57.5 to 4.5 times that of a high-energy scan for conventional DECT. Furthermore, the air kerma of bf-DECT could be reduced to less than that of conventional DECT, while obtaining the same figure of merit for the measurement of electron density and effective atomic number. Conclusions: The tube-loading and dose efficiencies of bf-DECT were considerably improved by sacrificing the quality of the noise level in the images of effective atomic number.

60 APPLIED LIFE SCIENCES↗

Scalable Asynchronous Domain Decomposition Solvers

We discuss how parallel implementations of linear iterative solvers generally alternate between phases of data exchange and phases of local computation. Increasingly large problem sizes and more heterogeneous compute architectures make load balancing and the design of low latency network interconnects that are able to satisfy the communication requirements of linear solvers very challenging tasks. In particular, global communication patterns such as inner products become increasingly limiting at scale. We explore the use of asynchronous communication based on one-sided Message Passing Interface primitives in the context of domain decomposition solvers. In particular, a scalable asynchronous two-level Schwarz method is presented. We discuss practical issues encountered in the development of a scalable solver and show experimental results obtained on a state-of-the-art supercomputer system that illustrate the benefits of asynchronous solvers in load balanced as well as load imbalanced scenarios. Using the novel method, we can observe speedups of up to four times over its classical synchronous equivalent.

97 MATHEMATICS AND COMPUTING↗

Improved load distribution in parallel sparse Cholesky factorization

Compared to the customary column-oriented approaches, block-oriented, distributed-memory sparse Cholesky factorization benefits from an asymptotic reduction in interprocessor communication volume and an asymptotic increase in the amount of concurrency that is exposed in the problem. Unfortunately, block-oriented approaches (specifically, the block fan-out method) have suffered from poor balance of the computational load. As a result, achieved performance can be quite low. This paper investigates the reasons for this load imbalance and proposes simple block mapping heuristics that dramatically improve it. The result is a roughly 20% increase in realized parallel factorization performance, as demonstrated by performance results from an Intel Paragon system. We have achieved performance of nearly 3.2 billion floating point operations per second with this technique on a 196-node Paragon system.

Rothberg, Edward↗

Parallel adaptive wavelet collocation method for PDEs

A parallel adaptive wavelet collocation method for solving a large class of Partial Differential Equations is presented. The parallelization is achieved by developing an asynchronous parallel wavelet transform, which allows one to perform parallel wavelet transform and derivative calculations with only one data synchronization at the highest level of resolution. The data are stored using tree-like structure with tree roots starting at a priori defined level of resolution. Both static and dynamic domain partitioning approaches are developed. For the dynamic domain partitioning, trees are considered to be the minimum quanta of data to be migrated between the processes. This allows fully automated and efficient handling of non-simply connected partitioning of a computational domain. Dynamic load balancing is achieved via domain repartitioning during the grid adaptation step and reassigning trees to the appropriate processes to ensure approximately the same number of grid points on each process. The parallel efficiency of the approach is discussed based on parallel adaptive wavelet-based Coherent Vortex Simulations of homogeneous turbulence with linear forcing at effective non-adaptive resolutions up to 2048{sup 3} using as many as 2048 CPU cores.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Dynamic load balancing for petascale quantum Monte Carlo applications: The Alias method

Diffusion Monte Carlo is the most accurate widely used Quantum Monte Carlo method for the electronic structure of materials, but it requires frequent load balancing or population redistribution steps to maintain efficiency and avoid accumulation of systematic errors on parallel machines. The load balancing step can be a significant factor affecting performance, and will become more important as the number of processing elements increases. We propose a new dynamic load balancing algorithm, the Alias Method, and evaluate it theoretically and empirically. An important feature of the new algorithm is that the load can be perfectly balanced with each process receiving at most one message. It is also optimal in the maximum size of messages received by any process. We also optimize its implementation to reduce network contention, a process facilitated by the low messaging requirement of the algorithm. Empirical results on the petaflop Cray XT Jaguar supercomputer at ORNL showing up to 30% improvement in performance on 120,000 cores. The load balancing algorithm may be straightforwardly implemented in existing codes. The algorithm may also be employed by any method with many near identical computational tasks that requires load balancing.

97 MATHEMATICS AND COMPUTING↗

Parallel FETI‐DP algorithm for efficient simulation of large‐scale EM problems

Summary An efficient parallelization of the dual‐primal finite‐element tearing and interconnecting (FETI‐DP) algorithm is presented for large‐scale electromagnetic simulations. As a nonoverlapping domain decomposition method, the FETI‐DP algorithm formulates a global interface problem, whose iterative solution is accelerated with a solution of a global corner problem. To achieve a good load balance for parallel computation, the original computational domain is decomposed into subdomains with similar sizes and shapes. The subdomains are then distributed to processors based on their close proximity to minimize inter‐processor communication. The parallel generalized minimal residual method, enhanced with the iterative classical Gram‐Schmidt orthogonalization scheme to reduce global communication, is adopted to solve the global interface problem with a fast convergence rate. The global corner‐related coarse problem is solved iteratively with a parallel communication‐avoiding biconjugate gradient stabilized method to minimize global communication, and its convergence is accelerated by a diagonal preconditioner constructed from the coarse system matrix. To alleviate neighboring communication overhead, the non‐blocking communication approach is employed in both generalized minimal residual and communication‐avoiding biconjugate gradient stabilized iterative solutions. Three numerical examples are presented to demonstrate the accuracy, scalability, and capability of the proposed parallel FETI‐DP algorithm for electromagnetic modeling of general objects and antenna arrays. Copyright © 2016 John Wiley & Sons, Ltd.

Zhang, Kedi↗

Randomized Cholesky Preconditioning for Graph Partitioning Applications

A graph is a mathematical representation of a network; we say it consists of a set of vertices, which are connected by edges. Graphs have numerous applications in various fields, as they can model all sorts of connections, processes, or relations. For example, graphs can model intricate transit systems or the human nervous system. However, graphs that are large or complicated become difficult to analyze. This is why there is an increased interest in the area of graph partitioning, reducing the size of the graph into multiple partitions. For example, partitions of a graph representing a social network might help identify clusters of friends or colleagues. Graph partitioning is also a widely used approach to load balancing in parallel computing. The partitioning of a graph is extremely useful to decompose the graph into smaller parts and allow for easier analysis. There are different ways to solve graph partitioning problems. For this work, we focus on a spectral partitioning method which forms a partition based upon the eigenvectors of the graph Laplacian (details presented in Acer, et. al.). This method uses the LOBPCG algorithm to compute these eigenvectors. LOBPCG can be accelerated by an operator called a preconditioner. For this internship, we evaluate a randomized Cholesky (rchol) preconditioner for its effectiveness on graph partitioning problems with LOBPCG. We compare it with two standard preconditioners: Jacobi and Incomplete Cholesky (ichol). This research was conducted from August to December 2021 in conjunction with Sandia National Laboratories.

97 MATHEMATICS AND COMPUTING↗

Spectral element methods: Algorithms and architectures

Spectral element methods are high-order weighted residual techniques for partial differential equations that combine the geometric flexibility of finite element methods with the rapid convergence of spectral techniques. Spectral element methods are described for the simulation of incompressible fluid flows, with special emphasis on implementation of spectral element techniques on medium-grained parallel processors. Two parallel architectures are considered: the first, a commercially available message-passing hypercube system; the second, a developmental reconfigurable architecture based on Geometry-Defining Processors. High parallel efficiency is obtained in hypercube spectral element computations, indicating that load balancing and communication issues can be successfully addressed by a high-order technique/medium-grained processor algorithm-architecture coupling.

Fischer, Paul↗

Spectral element methods - Algorithms and architectures

Spectral element methods are high-order weighted residual techniques for partial differential equations that combine the geometric flexibility of finite element methods with the rapid convergence of spectral techniques. Spectral element methods are described for the simulation of incompressible fluid flows, with special emphasis on implementation of spectral element techniques on medium-grained parallel processors. Two parallel architectures are considered; the first, a commercially available message-passing hypercube system; the second, a developmental reconfigurable architecture based on Geometry-Defining Processors. High parallel efficiency is obtained in hypercube spectral element computations, indicating that load balancing and communication issues can be successfully addressed by a high-order technique/medium-grained processor algorithm-architecture coupling.

Fischer, Paul↗

Dynamic Load Balancing Based on Constrained K-D Tree Decomposition for Parallel Particle Tracing

Particle tracing is a fundamental technique in flow field data visualization. In this work, we present a novel dynamic load balancing method for parallel particle tracing. Specifically, we employ a constrained k-d tree decomposition approach to dynamically redistribute tasks among processes. Each process is initially assigned a regularly partitioned block along with duplicated ghost layer under the memory limit. During particle tracing, the k-d tree decomposition is dynamically performed by constraining the cutting planes in the overlap range of duplicated data. This ensures that each process is reassigned particles as even as possible, and on the other hand the new assigned particles for a process always locate in its block. Result shows good load balance and high efficiency of our method.

Zhang, Jiang↗

Load Balancing Scientific Applications

The largest supercomputers have millions of independent processors, and concurrency levels are rapidly increasing. For ideal efficiency, developers of the simulations that run on these machines must ensure that computational work is evenly balanced among processors. Assigning work evenly is challenging because many large modern parallel codes simulate behavior of physical systems that evolve over time, and their workloads change over time. Furthermore, the cost of imbalanced load increases with scale because most large-scale scientific simulations today use a Single Program Multiple Data (SPMD) parallel programming model, and an increasing number of processors will wait for the slowest one at the synchronization points. To address load imbalance, many large-scale parallel applications use dynamic load balance algorithms to redistribute work evenly. The research objective of this dissertation is to develop methods to decide when and how to load balance the application, and to balance it effectively and affordably. We measure and evaluate the computational load of the application, and develop strategies to decide when and how to correct the imbalance. Depending on the simulation, a fast, local load balance algorithm may be suitable, or a more sophisticated and expensive algorithm may be required. We developed a model for comparison of load balance algorithms for a specific state of the simulation that enables the selection of a balancing algorithm that will minimize overall runtime.

97 MATHEMATICS AND COMPUTING↗

Automatic Cloud Bursting under FermiCloud

Cloud computing is changing the infrastructure upon which scientific computing depends from supercomputers and distributed computing clusters to a more elastic cloud-based structure. The service-oriented focus and elasticity of clouds can not only facilitate technology needs of emerging business but also shorten response time and reduce operational costs of traditional scientific applications. Fermi National Accelerator Laboratory (Fermilab) is currently in the process of building its own private cloud, FermiCloud, which allows the existing grid infrastructure to use dynamically provisioned resources on FermiCloud to accommodate increased but dynamic computation demand from scientists in the domains of High Energy Physics (HEP) and other research areas. Cloud infrastructure also allows to increase a private cloud’s resource capacity through “bursting” by borrowing or renting resources from other community or commercial clouds when needed. This paper introduces a joint project on building a cloud federation to support HEP applications between Fermi National Accelerator Laboratory and Korea Institution of Science and Technology Information, with technical contributions from the Illinois Institute of Technology. In particular, this paper presents two recent accomplishments of the joint project: (a) cloud bursting automation and (b) load balancer. Automatic cloud bursting allows computer resources to be dynamically reconfigured to meet users’ demands. The load balance algorithm which the cloud bursting depends on decides when and where new resources need to be allocated. Our preliminary prototyping and experiments have shown promising success, yet, they also have opened new challenges to be studied

Wu, Hao↗

Balancing Particle and Mesh Computation in a Particle-In-Cell Code

The XGC1 plasma microturbulence particle-in-cell simulation code has both particle-based and mesh-based computational kernels that dominate performance. Both of these are subject to load imbalances that can degrade performance and that evolve during a simulation. Each separately can be addressed adequately, but optimizing just for one can introduce significant load imbalances in the other, degrading overall performance. A technique has been developed based on Golden Section Search that minimizes wallclock time given prior information on wallclock time, and on current particle distribution and mesh cost per cell, and also adapts to evolution in load imbalance in both particle and mesh work. In problems of interest this doubled the performance on full system runs on the XK7 at the Oak Ridge Leadership Computing Facility compared to load balancing only one of the kernels.

Worley, Patrick H↗

Improved Program For Calculation Of Heat-Load Multiplier

PRM1940 computer program computes heat-load multiplier for use in Power Balance Model (PBM) computer program which calculates hundreds of operating parameters of main engine of space shuttle from relatively few measurement data. PRM1940 is stand-alone program which incorporates only those PBM calculations necessary to compute heat-load multiplier. Developed to accelerate and partly automate calculation of heat-load multiplier. Although programs specific to space shuttle application, also of interest to engineers concerned with monitoring of conditions in turbines, chemical-processing plants, and other high-temperature flow machinery.

D'Valentine, Mark↗

Dynamic Load Balancing For Grid Partitioning on a SP-2 Multiprocessor: A Framework

Computational requirements of full scale computational fluid dynamics change as computation progresses on a parallel machine. The change in computational intensity causes workload imbalance of processors, which in turn requires a large amount of data movement at runtime. If parallel CFD is to be successful on a parallel or massively parallel machine, balancing of the runtime load is indispensable. Here a framework is presented for dynamic load balancing for CFD applications, called Jove. One processor is designated as a decision maker Jove while others are assigned to computational fluid dynamics. Processors running CFD send flags to Jove in a predetermined number of iterations to initiate load balancing. Jove starts working on load balancing while other processors continue working with the current data and load distribution. Jove goes through several steps to decide if the new data should be taken, including preliminary evaluate, partition, processor reassignment, cost evaluation, and decision. Jove running on a single IBM SP2 node has been completely implemented. Preliminary experimental results show that the Jove approach to dynamic load balancing can be effective for full scale grid partitioning on the target machine IBM SP2.

Sohn, Andrew↗

Dynamic Load Balancing for Grid Partitioning on a SP-2 Multiprocessor: A Framework

Computational requirements of full scale computational fluid dynamics change as computation progresses on a parallel machine. The change in computational intensity causes workload imbalance of processors, which in turn requires a large amount of data movement at runtime. If parallel CFD is to be successful on a parallel or massively parallel machine, balancing of the runtime load is indispensable. Here a framework is presented for dynamic load balancing for CFD applications, called Jove. One processor is designated as a decision maker Jove while others are assigned to computational fluid dynamics. Processors running CFD send flags to Jove in a predetermined number of iterations to initiate load balancing. Jove starts working on load balancing while other processors continue working with the current data and load distribution. Jove goes through several steps to decide if the new data should be taken, including preliminary evaluate, partition, processor reassignment, cost evaluation, and decision. Jove running on a single EBM SP2 node has been completely implemented. Preliminary experimental results show that the Jove approach to dynamic load balancing can be effective for full scale grid partitioning on the target machine IBM SP2.

Sohn, Andrew↗

I/O load balancing for big data HPC applications

High Performance Computing (HPC) big data problems require efficient distributed storage systems. However, at scale, such storage systems often experience load imbalance and resource contention due to two factors: the bursty nature of scientific application I/O; and the complex I/O path that is without centralized arbitration and control. For example, the extant Lustre parallel file system-that supports many HPC centers-comprises numerous components connected via custom network topologies, and serves varying demands of a large number of users and applications. Consequently, some storage servers can be more loaded than others, which creates bottlenecks and reduces overall application I/O performance. Existing solutions typically focus on per application load balancing, and thus are not as effective given their lack of a global view of the system. In this paper, we propose a data-driven approach to load balance the I/O servers at scale, targeted at Lustre deployments. To this end, we design a global mapper on Lustre Metadata Server, which gathers runtime statistics from key storage components on the I/O path, and applies Markov chain modeling and a minimum-cost maximum-flow algorithm to decide where data should be placed. Evaluation using a realistic system simulator and a real setup shows that our approach yields better load balancing, which in turn can improve end-to-end performance.

Paul, Arnab K.↗