Compiler Technology for Scalable Servers

本研究旨在探讨提高网际网路伺服器程式效能与可扩充性的编译器技术。服务器具有一组不同于编译器作者过去所关注的计算的特征。服务器通常使用许多短期线程来管理来自多个客户端的I/O边界连接，而不是使用适度数量的长时间线程来执行受计算机限制的计算。这些新特性使系统以前不太相关的方面变得非常重要，例如线程创建开销和线程堆栈开销。它们还提供了编译器可以利用的机会来改进内存管理。这项研究将通过自动将多线程代码转换为事件驱动的形式来攻击线程开销。事件驱动形式的代码包含一个事件循环，该循环反复阻塞等待来自任何当前连接的下一个事件。然后它调用一个处理程序来执行适当的操作，然后返回到事件循环。这种转换将消除线程创建和管理开销。编译器将通过分析程序来识别那些使用时间绑定到单个线程的对象，然后对这些对象使用专门的内存管理算法，从而降低内存管理开销。

This research investigates compiler techniques for improving theperformance and scalability of Internet server programs. Servers havea different set of characteristics from the computations that compilerwriters have focused on in the past. Instead of executing acompute-bound computation with a modest number of long-lived threads,servers typically use many short-lived threads to manage I/O boundconnections from multiple clients. These new characteristics place apremium on previously less relevant aspects of the system such as thethread creation overhead and thread stack overhead. They also provideopportunities that the compiler can exploit to improve the memorymanagement. The research will attack thread overhead by automaticallytransforming multithreaded code to event-driven form. Code inevent-driven form contains an event loop that repeatedly blockswaiting for the next event from any of the current connections. Itthen invokes a handler that executes the appropriate action, thenreturns to the event loop. This transformation will eliminate threadcreation and management overhead. The compiler will attack memorymanagement overhead by analyzing the program to identify objects whoselifetimes are tied to individual threads, then using specializedmemory management algorithms for these objects.