高度可擴充性DVB-MHP 平台上的軟硬體協同設計

Abstract

多媒體家用平台(MHP)是由Digital Video Broadcasting(DVB)所提 出，作為互動電視家用娛樂平台上的中介軟體公開標準，這個平台使用 Java 為主要的程式語言，由Java 虛擬機器(VM)負責程式的運作執行。採 用軟硬體協同設計的方式能讓Java 虛擬機器具有高度的可擴充性，功能 也強大許多，但仍舊會受限於Java 語言本身的效率不彰;而傳統的動態編 碼最佳化雖然可以利用一些執行時期所得的資訊來加速系統，但對於嵌入 式系統來說，這個代價是十分昂貴的。因此，在這篇論文中，我們提出了 一個新的動態編碼最佳化演算法，用軟硬體協同設計的方式使這類系統的 整體效能大大的提升，並且更加的省電。我們將這樣的想法實作在Java Optimized Processor(JOP)上，並且在Xilinx 的Spartan-3 發展板上模 擬執行，實驗結果顯示我們所提出的這套架構在整體上可增進13.8%的速 度;在省電方面，也分別可以減少10.4%的微指令執行週期以及11.1%的外 部記憶體存取。

Multimedia Home Platform (MHP) is the open middleware system for interactive television and related interactive home entertainment designed by the Digital Video Broadcasting (DVB) project. They use Java as the common programming language and embed the Java Virtual Machine (VM) that provides a stable and cross-platform java runtime environment in the system software layer. A hardware/software co-design approach makes Java VM more flexible and powerful, but it still suffers from the inefficiency of java system. Typical dynamic code optimization can save method lookup and constant pool searching time using the runtime information known in the first time we execute it. However, in such kind of embedded system, it is very expensive due to the overhead of external memory modification. In this thesis, we propose a new hardware/software co-design dynamic code optimization schema for this kind of approach that can significantly improve the efficiency of Java program execution. By analyzing the execution frequency of Java code segment, we can dynamically decide if the dynamic code optimization is needed. This approach can also cut down the power consumption with less microcode execution cycles and less external memory access. We implement this architecture on Java Optimized Processor (JOP) and simulate on Xilinx Spartan-3 developing board. Experiment Result shows that this proposed dynamic code optimization schema for Java VM hardware/software co-design of DVB-MHP terminal has 13.8% average speedup, 10.4% less microcode execution cycles and 11.1% less external memory access than the original system.