Shared memory and distributed shared memory systems: A survey

Kavi, Krishna M.; Kim, Hyong-Shik; Liu, Ben; Hurson, Ali R.

doi:10.1016/s0065-2458(00)80004-2

Cited by 3 publications

(4 citation statements)

References 43 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This can become a bottleneck on distributed memory systems, where variables need to be passed around between processors. A drawback with shared memory, however, is that it suffers from poor scalability [91]. Distributed memory could thus be advantageous when simulating a large number of relatively large sub-models.…”

Section: Methodsmentioning

confidence: 99%

Distributed System Simulation Methods : For Model-Based Product Development

Braun

2015

Linköping Studies in Science and Technology. Dissertations

View full text Add to dashboard Cite

With the current trend towards multi-core processors and computer grids, parallelism in system simulation is becoming inevitable. Performing multiple tasks concurrently can save both time and money. However, this also puts higher demands on simulation software. This thesis investigates how simulation-based product development can benefit from distributed simulation.One parallelization method is to cut apart models for simulation with distributed solvers by using time delays. The transmission line element method (TLM) is used, which eliminates numerical errors by introducing physically motivated time delays. Results show that linear speed-ups can be obtained for large models with balanced workloads. Different task scheduling techniques are tested. It is found that a fork-join algorithm with implicit synchronization performs best for models with a large total workload. On the other hand, a task-pool implementation with lock-based synchronization works better with smaller workloads.The distributed solver method virtually equals co-simulation between different simulation tools. Co-simulation also induces time delays at the interface variables, which makes TLM a useful method. An emerging standard for coupling of simulation tools is the Functional Mock-up Interface (FMI). Experiments show that FMI works well together with TLM, even when connecting different simulation domains such as system simulation and multi-body mechanics. Combining FMI with TLM enables a framework for tool-independent co-simulation with distributed solvers and time steps.When introducing distributed solvers, it can be suitable to maintain the same structure that was used in the modelling process. A popular object-oriented equation-based modelling language is Modelica. Instead of flattening the model for simulation with a centralized solver, TLM makes it possible to use different solvers for different objects in the model. Two approaches are investigated. First, Modelica support is implemented in the Hopsan TLM tool with a code generator. Second, Modelica models are imported to Hopsan through the FMI interface. Both methods are feasible. While the first requires less overhead i work, the second makes it possible to mix models from different Modelica tools and to use all the features of each individual tool. Depending on the number of TLM elements and their locations, model structures with different granularities can be achieved.Parallelism can be introduced at different levels. The distributed solver approach results in parallelism at the model-level. It is also possible to parallelize equation solvers, or to run several simulations in parallel. The latter is especially useful for simulation-based design optimization. An algorithm for profile-based scheduling of parallel optimization jobs is proposed that takes parallelism at both model-level and optimization level into account. Parallel optimizations can then be executed on either homogeneous or heterogeneous computer grids. This makes it possible to utilize the large amount of unused co...

show abstract

Section: Methodsmentioning

confidence: 99%

Distributed System Simulation Methods : For Model-Based Product Development

Braun

2015

Linköping Studies in Science and Technology. Dissertations

View full text Add to dashboard Cite

show abstract

“…In distributed memory MIMD, processors access local memory directly, and use messages system to access remote data in other processors (Kavi et al 2000, Jordan & Alaghband 2003. MPI is a parallel programing language designed for distributed memory MIMD which can also run on shared memory MIMD computers (El-Nashar 2011, Jordan & Alaghband 2003.…”

Section: Shared Memory and Distributed Memory Mimdmentioning

confidence: 99%

“…OpenMP is a parallel programing language designed for shared memory MIMD (Jordan & Alaghband 2003). The shared memory also improves the ability of parallelizing compilers, standard operating systems, resource management and incremental performance tuning of applications (Kavi et al 2000). However, there is a cost associated with parallelism using OpenMp.…”

Section: Shared Memory and Distributed Memory Mimdmentioning

confidence: 99%

“…This cost represents the overhead of threads management including the time to create, start, and stop threads, the time to assign the work to the threads, and the time spent for synchronization (Chapman et al 2007). In distributed memory MIMD, processors access local memory directly, and use messages system to access remote data in other processors (Kavi et al 2000, Jordan & Alaghband 2003. MPI is a parallel programing language designed for distributed memory MIMD which can also run on shared memory MIMD computers (El-Nashar 2011, Jordan & Alaghband 2003.…”

Section: Shared Memory and Distributed Memory Mimdmentioning

confidence: 99%

See 1 more Smart Citation

High Performance Parallel Sort for Shared and Distributed Memory Mimd

Alghamdi¹,

Alaghband²

2019

Proceedings of the 16th International Conference on Applied Computing 2019

View full text Add to dashboard Cite

We present four high performance hybrid sorting methods developed for various parallel platforms: shared memory multiprocessors, distributed multiprocessors, and clusters taking advantage of existence of both shared and distributed memory. Merge sort, known for its stability, is used to design several of our algorithms. We improve its parallel performance by combining it with Quicksort. We present two models designed for shared memory MIMD (OpenMP): (a) a non-recursive Merge sort and (b) a hybrid Quicksort and Merge sort. The third model presented is designed for distributed memory MIMD (MPI) using a hybrid Quicksort and Merge sort. Our fourth model is designed to take advantage of the shared memory within individual nodes of today's cluster systems, and to eliminate all internal data transfers between different nodes, Our model implements a one-step MSD-Radix to distribute data in ten packets (MPI) while parallel cores of each node use Quicksort to sort their data partitions sequentially then merge and sort them in parallel employing the OpenMp. The performances of all developed models outperform the baseline performance. Hybrid Quicksort and Merge sort outperformed Hybrid Memory Parallel Merge Sort using Hybrid MSD-Radix and Quicksort in Cluster Platforms when sorting small size data, but with larger data the speedup of Hybrid Memory Parallel Merge Sort Using Hybrid MSD-Radix and Quicksort in Cluster Platforms becomes bigger and it keeps improving. The speedup of Distributed Memory Parallel Hybrid Quicksort and Merge Sort is the best.

show abstract

Efficiency in Parallel Computing of FDS Model for Compartment Fire Simulation: Shared Memory System

민¹,

김²

2021

Fire Sci. Eng.

View full text Add to dashboard Cite

This study evaluates the computational efficiency based on the parallel processing mode and domain decomposition method of the FDS model to enhance the computational performance of fire simulation. A single compartment of dimensions 12.0 m × 3.8 m × 3.0 m is considered along with a rectangular fire source (0.4 m × 0.4 m) fueled by n-Heptane. The computational domain was divided into 136,000 cells forming a grid size of 0.1 m, and the computational efficiency for each calculation was evaluated by the wall clock time for a simulation time of 300 s using a computational framework with 24 cores of a single CPU and a 256 GB shared memory system. The MPI and hybrid mode in FDS parallel offers a greater speed-up capability than the OpenMP mode, and the domain decomposition method used greatly affects the computational efficiency. The maximum speed-up with the OpenMP mode was less than 1.5 for a single computational domain, which indicates that there is an optimal condition for thread assignment and domain decomposition in the OpenMP mode. The present study is expected to contribute toward obtaining effective fire simulation results with limited computing power and time in fire protection engineering.

show abstract

Shared memory and distributed shared memory systems: A survey

Cited by 3 publications

References 43 publications

Distributed System Simulation Methods : For Model-Based Product Development

Distributed System Simulation Methods : For Model-Based Product Development

High Performance Parallel Sort for Shared and Distributed Memory Mimd

Efficiency in Parallel Computing of FDS Model for Compartment Fire Simulation: Shared Memory System

Contact Info

Product

Resources

About