With the surge of big data applications and the worsening of the memory-wall problem,the memory system,instead of the computing unit,becomes the commonly recognized major concern of computing.However,this“memorycent...With the surge of big data applications and the worsening of the memory-wall problem,the memory system,instead of the computing unit,becomes the commonly recognized major concern of computing.However,this“memorycentric”common understanding has a humble beginning.More than three decades ago,the memory-bounded speedup model is the first model recognizing memory as the bound of computing and provided a general bound of speedup and a computing-memory trade-off formulation.The memory-bounded model was well received even by then.It was immediately introduced in several advanced computer architecture and parallel computing textbooks in the 1990’s as a must-know for scalable computing.These include Prof.Kai Hwang’s book“Scalable Parallel Computing”in which he introduced the memory-bounded speedup model as the Sun-Ni’s Law,parallel with the Amdahl’s Law and the Gustafson’s Law.Through the years,the impacts of this model have grown far beyond parallel processing and into the fundamental of computing.In this article,we revisit the memory-bounded speedup model and discuss its progress and impacts in depth to make a unique contribution to this special issue,to stimulate new solutions for big data applications,and to promote data-centric thinking and rethinking.展开更多
A design of a new heterogeneous code for LBM simulations is proposed.By heterogeneous computing wemean a collaborative computation on CPU and GPU,which is characterized by the following features:the data is distribute...A design of a new heterogeneous code for LBM simulations is proposed.By heterogeneous computing wemean a collaborative computation on CPU and GPU,which is characterized by the following features:the data is distributed between CPU and GPU memory spaces taking advantage of both parallel hierarchies;the capabilities of both SIMT GPU and SIMD GPU parallelization are used for calculations;the algorithms in use efficiently conceal the CPU-GPU data exchange;the subdivision of the computing task is performed with an account for the strong points of both processing units:high performance of GPU,low latency,and advanced memory hierarchy of CPU.This code is a continuation of our work in the development of LRnLA codes for LBM.Previous LRnLA codes had good efficiency both for CPU and GPU computing,and allowed GPU simulation performed on data stored in CPU RAM without performance loss on CPU-GPU data transfer.In the new code,we use methods and instruments that can be flexibly adapted to GPU and CPU instruction sets.We present the theoretical study of the performance of the proposed code and suggest implementation techniques.The bottlenecks are identified.As a result,we conclude that larger problems can be simulated with higher efficiency in the heterogeneous system.展开更多
基金supported in part by the U.S.National Science Foundation under Grant Nos.CCF-2029014 and CCF-2008907.
文摘With the surge of big data applications and the worsening of the memory-wall problem,the memory system,instead of the computing unit,becomes the commonly recognized major concern of computing.However,this“memorycentric”common understanding has a humble beginning.More than three decades ago,the memory-bounded speedup model is the first model recognizing memory as the bound of computing and provided a general bound of speedup and a computing-memory trade-off formulation.The memory-bounded model was well received even by then.It was immediately introduced in several advanced computer architecture and parallel computing textbooks in the 1990’s as a must-know for scalable computing.These include Prof.Kai Hwang’s book“Scalable Parallel Computing”in which he introduced the memory-bounded speedup model as the Sun-Ni’s Law,parallel with the Amdahl’s Law and the Gustafson’s Law.Through the years,the impacts of this model have grown far beyond parallel processing and into the fundamental of computing.In this article,we revisit the memory-bounded speedup model and discuss its progress and impacts in depth to make a unique contribution to this special issue,to stimulate new solutions for big data applications,and to promote data-centric thinking and rethinking.
基金supported by Russian Science Foundation,grant#18-71-10004.
文摘A design of a new heterogeneous code for LBM simulations is proposed.By heterogeneous computing wemean a collaborative computation on CPU and GPU,which is characterized by the following features:the data is distributed between CPU and GPU memory spaces taking advantage of both parallel hierarchies;the capabilities of both SIMT GPU and SIMD GPU parallelization are used for calculations;the algorithms in use efficiently conceal the CPU-GPU data exchange;the subdivision of the computing task is performed with an account for the strong points of both processing units:high performance of GPU,low latency,and advanced memory hierarchy of CPU.This code is a continuation of our work in the development of LRnLA codes for LBM.Previous LRnLA codes had good efficiency both for CPU and GPU computing,and allowed GPU simulation performed on data stored in CPU RAM without performance loss on CPU-GPU data transfer.In the new code,we use methods and instruments that can be flexibly adapted to GPU and CPU instruction sets.We present the theoretical study of the performance of the proposed code and suggest implementation techniques.The bottlenecks are identified.As a result,we conclude that larger problems can be simulated with higher efficiency in the heterogeneous system.
