查詢結果分析
相關文獻
- Performance Comparison of Four Interconnection Networks for Multiprocessor Systems
- Design and Implementation of a Parallelism Detector for CONVEX SPP-1000 System
- 地方公共服務績效比較評量之探討--標竿學習策略的觀點
- An Introduction to a PC Cluster with Diskless Slave Nodes
- Using a PC Cluster for High-Performance Computing and Applications
- 由整合異質性網路來實作--具彈性的無縫漫遊網路
- 一個抽象多處理機系統效能之模式與分析
- 日本股票市場的過度反應、特徵值、風險與股票報酬的關係
- Design and Implementation of A Dynamic Data-Driven Multiprocessor System on Transputer Networks
- 以多處理機系統作U 型機械人路徑控制及動態畫研究
頁籤選單縮合
| 題 名 | Performance Comparison of Four Interconnection Networks for Multiprocessor Systems=四種多處理機系統下連結方式的績效比較 |
|---|---|
| 作 者 | 謝忠健; | 書刊名 | 大同學報 |
| 卷 期 | 23 1993.11[民82.11] |
| 頁 次 | 頁223-232 |
| 分類號 | 448.595 |
| 關鍵詞 | 多處理機系統; 連結方式; 績效比較; |
| 語 文 | 英文(English) |
| 中文摘要 | 本篇論文利用實驗數據比較四種一般用於多處理機系統下連結方式的績效。該四 種連結是全連結、立方連接、平面網狀連結及共用記憶體式連結。 我們用以評估績效的方法是把應用程式的每個指令,經過時隙預留排序法則的安排後,求取 在每種連結之下的執行時間。時隙預留排序法則在排序時為使每一指令在多處理機系統下能 最快的被執行,考慮了資料相關、處理機管道效應及網路連結等因素,以找出最合適的處理 機及最合適的時隙。 經實驗數據發現全連結的效果最好,但卻是最昂貴;在硬體成本的考量下,平面網狀連結是 最好的選擇;而共同記憶體方式由於易於實現而廣為採用,但在績效上卻不如預期的好。 |
| 英文摘要 | This paper use empirical data to compare the performance of four general used interconnections for multiprocessor systems. The four connections are full connection, cube connection, mesh connection and shared memory. The approach we use to evaluate the performance of these four interconnections is to schedule the cited application programs at instruction level. The scheduler, SR-mapper, is a fine grain mapping strategy which assigns each instruction of the application programs to a most suitable processor taking data dependence between instructions, pipeline effect of each processor, and communication cost between processors into consideration. The results show that full connection is the best of all as expected. However if the hardware cost is the primary concern, mesh connection will be the one to use. Shared memory is not a good choice in general. |
本系統中英文摘要資訊取自各篇刊載內容。