What's new in the Project

• August 2000: the JCOD protype is demonstrated in Japan
• August 1999: the JCOD Compiler Project pages are created

Overview of the project

The JCOD project (Java Compiler On Demand) consists in developing a dynamic compile server technology for embedded systems applications. Such applications would run on a device (the Target) adhering to the following scenario:

• The Embedded Java application is downloaded from an application server into the Target system. The target and the application server can be connected to the Internet or any sort of company Intranet. The Target system runs a Java Virtual Machine that is the CHAI VM from Hewlett-Packard, augmented with new capabilities enabling it to support Compile-On-Demand of the application byte code.

• The goal of the compile-on-demand framework is to only compile the most frequently executed code. In order to know which code is most frequently executed, the VM is augmented with profiling capabilities.

• As the application is running, the target enhanced VM dynamically profiles the executed code. When the execution cost of a Java method becomes predominant, the JVM requests to a compile server to compile it.

• A compile server listens to the network for such compile requests, processes the request and sends a reply to the target when completed. To process a compile request, it invokes a compiler that generates native code. This native code is encoded in a specific "network load-able object format".

• Once the compiler has generated the native code, the compile server sends the object format it to the requesting target. Upon reply (receiving the compiled code), the target device installs the native code, relying on the services of a Dynamic Compiler API. On the next invocation of the corresponding application method the JVM will automatically switch to the native code.

• In order to encourage the wide spread of this technology to open a third party market, and to speed up its deployment to many embedded microprocessors, the specifications of the compiler back-end, that allow to generate code for a particular processor, will be made public.

• During this overall dynamic compile on demand process, the system behavior can be observed through the JCOD User Access, which provides a user interface allowing monitoring and tracing of the system activity.

• The compiler architecture includes a microprocessor independent front-end that interfaces with a microprocessor dependent back-end (a code generator). The compiler consists of a single front-end and as many back-ends as there are device (microprocessor) types. We will publish the specifications of the back-end component so that anyone can implement a back-end of its own for any microprocessor. The back-end specifications will consist of a Java API specification for the objects generated by the front-end, with a description of the methods that the back-end object must implement in order to generate native code. As long as this back-end library is compliant with the specifications, it can generate code for any microprocessor. Therefore this Java specs can be implemented by any third party.

Architecture

The project includes three major functional components:

• a Target Enhanced Java Virtual Machine that is a client of the Compile Server.
• a Compile Server which invokes a compiler to translate the bytecode into native code.
• a JCOD User Access, which logs the related activity and provides developers and system administrator with means for starting/shutting down the system, for tracing of errors, and viewing performance measurements.

The Targets communicates with the server over the network through a dedicated protocol. This protocol is named the Java Compile-On-Demand Protocol (JCODP). Therefore in the sequel of this document, we will refer to the components as the JCOD Compile server, the JCOD User Access and the JCOD Target Client JVM.

The JCOD User Access communicates with the Compile Server using the HTTP Internet protocol, as a Web browser is the basic tool that is used by users to access information stored on the server.

Si vous avez un probleme avec cette page, svp contacter le ouaibe-master.

Last modified : Friday, 15-Sep-2000 17:49:02 CEST