http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
데이타흐름/제어흐름 혼합계산 모델을 기초로한 다중스레드 컴퓨터 구조의 동적 캐쉬충전 기법
노권형,황대준 成均館大學校 科學技術硏究所 1992 論文集 Vol.43 No.2
In most multithreaded, and multithreaded computer architectures, the cache refill has been performed based on the locality of memory references. The scheme, however, would not be effective anymore in the multithreaded architecture because context switching to another one will inevitably require initial cache misses every time the switched context gets mapped onto the target processor for its execution. Under this situation, processor utilization begins to sharply drop and the processing power of the processor gets severly damaged. Unlike von Neumann computational model, in the dataflow/control flow hybrid computational model the information of enabled threads becomes available before their execution. With Based on the prior information enabled threads are to be scheduled and executed by the target processor, keeping the initial cache miss ratio nearly zero. And this result was confirmed by the simulation over several benchmarks running on the MACAW, which has been developing since the very early of 1992 for evaluating architectural features of MPP nodes.
다중스레드 프로세서 구조의 캐쉬 충전 기법에 관한 연구
노권형(Kwon-Hyung Rho),옥은정(Eun-Jung Yuk),황대준(Dae-Joon Hwang) 한국정보과학회 1993 한국정보과학회 학술발표논문집 Vol.20 No.1
캐쉬 기억 장치의 성능을 향상 시키기 위한 다양한 캐쉬 충전(cache refill) 기법들이 제안되었으나 데이타 흐름/제어 흐름의 혼합형 계산모델을 기반으로 하는 다중스레트 프로세서 구조에 적합한 캐쉬 충전 기법에 관한 연구는 아직 미진한 가운데 있다. 본 논문에서는 기존의 폰 노이만 프로세서 구조에서 적응되어온 지역성을 기반으로 하는 캐쉬 충전 기법들이 다중스레드 컴퓨터 구조에서는 적합하지 않음을 밝히고, 다중스레드 컴퓨터 구조에 적합한 컨택스트 기반(context-based) 캐쉬 충전 기법을 제안한다.
대단위 병렬컴퓨터 시스템 노드구조 설계를 위한 워크벤치(SKYLAB)의 개발
윤현진,육은정,노권형,황대준 成均館大學校 科學技術硏究所 1993 論文集 Vol.44 No.1
Building the simulation environment of massively parallel computer systems has been recognized as a better approach over the experimentation on real machines to study extensively their architectural behaviours at a minimum cost. In this paper, we introduce a parallel processing workbench SKYLAB being implemented on DE C5000/25 and SPARCI workstations for design of node architectures featuring optimal organization for massively parallel processing. For the through analysis of the features of architectures in mind, the SKYLAB comes with thress major components: the compilerbackend translating single-threaded codes into their equivalent multithreaded codes, the node architecture emulator incopperating target CPU for program execution, and the graphic use interface for easy interpretation and expedating the analysis of the statitistics from the simulation runs. The design experience of the node architecture of the DAVRID multithreaded parallel processing computer using the SKYLAB examplifies how much effictively to make experiment on parallel simulations using this integrated toolset.