透過直接記憶體通道對應技術連結Java虛擬機及作業系統核心間資訊之效能分析工具

近年來，隨著嵌入式系統產品的大量普及化，對於系統效能的需求也逐年受到重視，但既有的效能分析程式並無法有效滿足開發人員進行各種不同的效能分析面向。我們發現若只繼續著重於應用程式面的評估和測試，將面臨著無法分辨是應用程式或是系統架構上所造成的效能瓶頸，因此我們將觀測系統設計的重點從應用程式面擴大到作業系統並和底層的硬體做連結，來進行全面系統性的效能分析。本篇論文提出使用直接記憶體通道的資訊對應技術來克服傳統上無法跨越虛擬機層次的效能分析問題。直接記憶體通道的方法可以有效降低在虛擬機執行過程中的額外效能負擔從數萬個時鐘周期降低到60至200左右的時鐘周期。最後提供一符號對應方法來提供事後完整的效能分析報告。

關鍵字

效能分析

並列摘要

Embedded system products become more and more popular in recent years. Performance analysis also plays an important role in hardware and software integration. But so many performance analysis tools are not sufficient with profiling in different system layers. The problem make programmers could not find the bottleneck between software, OS and hardware. This thesis proposes a system-wide profiling tool by using direct memory tunnel and data remapping technique to conquer the traditional problems. Direct memory tunnel provides a low-overhead that is lower than two hundred clock cycle instrument in virtual machines. Also this thesis provides a fully performance analysis report with symbol correction in virtual machine.

並列關鍵字

Vertical Profiling

參考文獻

[15] H. Mousa, C. Krintz, L. Youseff, and R. Wolski, “VIProf: Vertically Integrated Full-System Performance Profiler,”. IPDPS 2007. IEEE International, pp. 1–6

[18] “DIOTA

[30] Peter F. Sweeney, Matthias Hauswirth, and Amer Diwan,

[1] “Android Developers,”