應用於嵌入式異質多核心平台之爪哇虛擬機器

Abstract

在半導體製程的限制下，為了滿足對計算機系統日益增加的效能需求，工業界逐漸 傾向以增加處理器核心的方式來達到目的。而為了滿足中高階嵌入式系統的效能需求與 功率消耗上的限制，使用異質多核心作為嵌入式系統硬體平台的架構有越來越普遍的趨 勢。 從另一方面來看，現今通訊相關產業快速發展，資訊家電、個人數位助理 (PDA) 及 各種中高階電子裝置需求急速增加。因為各種產品的作業系統環境和執行平台差異性很 大，現在有許多廠商使用Java 語言來撰寫應用程式，發揮Java 語言跨平台的特性。為 了要執行Java 程式，必須在這些嵌入式系統上執行Java 虛擬機器。 然而在現階段，尚未存在完整適用於嵌入式異質多核心系統之 JVM 整合方案。其 VM 本身的組件及JIT 編譯器動態產生的原生碼必須能夠並行且平行被執行，且能依據 程式行為，利用異質多核心中不同核心的能力加速。 本計劃將以三年時間，探討實做一個完整支援嵌入式異質多核心系統的 JVM 整合 方案相關的各項重要工作及研究議題，可分成以下三個階段: 1. 移植及針對系統中各個處理器優化 JVM 及JIT 編譯器。 2. 探討 JVM 中的並行性(concurrency)與平行性(parallelism)。 3. 整合自動工作分派與排程機制。 預期將在計劃完成時，提出上述各項工作之解決方案以及系統整合之解決方法。此 外，也將實作整合各項技術之Embedded JVM

In order to adapt to the demanding performance requirements for computing systems, it becomes a common practice to put several processor cores on a chip. These cores are usually heterogeneous for middle/high-level embedded systems. Speed and power consumption are also crucial issues in these systems. In another respect, because of the rapid development of the communication industry, the market for Information Appliance (IA), Personal Digital Assistants (PDA), and many electronic devices is expanding rapidly. Since the operating systems and execution platforms for these products vary significantly, many enterprises to develop applications with the Java language due to its platform independency. Once a JVM (Java virtual machine) is ported to a new platform, all Java applications are available there. However, a complete and integrated JVM for embedded heterogeneous multi-core systems does not exist yet. The components of such a JVM and the compiled code produced by the JIT compiler inside the JVM should be able to be executed in parallel in order to achieve high performance.. In the coming three years, our project will focus on the research design and implementation of a complete and integrated JVM for embedded heterogeneous multi-core systems. Our project is divided into the following three stages: 1. Porting and optimizing JVM (including the JIT compiler inside the JVM ) for heterogeneous multi-core embedded systems. 2. Exploiting concurrency and parallelism in the multi-core systems. 3. Integrating an automatic task partitioning and scheduling mechanism. We will propose the solutions during the above stages. Our final products include various related techniques and a general and complete JVM for heterogeneous multi-core embedded systems.