Using Code Shipping to Optimize Distributed System Execution

Richard SchlichtingCCR-9972192代码从分布式系统中的一台机器移动到另一台机器通常被用作获得功能或性能优势的一种方式。 这个项目的重点是使用代码传送来优化分布式系统中的执行性能。 代码运送涉及从现有程序静态地提取代码片段，从片段构造以诸如Java的语言编写的可执行包，以及将包运送到另一机器以供执行。 代码传送的一个示例是实现对远程过程调用（RPC）的调用折叠，其中从客户端到服务器的多个RPC被折叠成单个RPC。 执行调用的程序片段（包括任何中间语句）在编译时被提取并生成为一个包，然后在运行时作为剩余RPC中的一个参数运送到服务器并在那里执行。 因此，最初需要网络传输的对服务器操作的调用现在是本地的。 与以前的方法相比，代码运送涉及从常规源代码自动生成包裹和其他软件组件，通常涉及更细粒度的片段，并强调性能优化而不是功能增强。 正在调查的问题范围从确定代码中的优化可能性到开发适当的转换技术。 我们的调查是基于务实的方法来解决这个问题，与正在做的工作的背景下，建立一个原型代码航运系统

Richard SchlichtingCCR-9972192The movement of code from one machine in a distributed system to another has often been used as a way to gain a functionality or performance advantage. This project focuses on using code shipping to optimize execution performance in distributed systems. Code shipping involves statically extracting code fragments form an existing program, constructing executable parcels written in a language such as Java from the fragments, and shipping the parcels to another machine for execution. An example use of code shipping is to implement call folding to remote procedure calls (RPCs), where multiple RPCs from a client to a server are folded into a single RPC. The program fragment that performs the calls, including any intervening statements, is extracted and generated as a parcel at compile time, and then shipped to the server at run time as an argument in the single remaining RPC and executed there. Thus, the calls to the server operations that originally required network transmissions are now local. In contrast to previous approached, code shipping involves automatic generation of parcels and other software components from regular source code, typically involves finer-grain fragments, and emphasizes performance optimization rather than function enhancement.The goal of this project is to explore code shipping as a technique for optimizing execution in distributed systems. The issues being investigated range from identifying optimization possibilities in the code to developing appropriate transformation techniques. Our investigations are based on pragmatic approached to the problem, with the work being done in the context of building a prototype code shipping sys