一种混合型运行时信息分析方法

A Mixed Runtime Information Analysis Methed

为适应新一代广播电视业务的需求,Java技术引起了机顶盒厂商的广泛关注。针对嵌入式Java执行效率不高的问题,本文提出一种采样与计数相结合的混合型运行时信息分析方法。通过监控方法调用堆栈的出栈入栈操作和回边事件,对Java方法调用和循环进行计数,并统计得到Java方法的执行时间。基于此提出了一种热点判定模型,并引入基于运行时信息记录引导的热点预判机制。基于J2MECDC在一款MIPS平台上对该方法进行仿真,CaffeineMark测试结果显示,采用历史信息预判时,整体性能能够提高2%,循环性能测试可提高6%。

Java technology was introduced to the software architecture of STB to support the new generation radio and television services.With the bad execution efficiency of embedded java,a hybrid run-time information profiling algorithm using sampling and counting was proposed.By monitoring the operations of the method stack,the invoke events and backwards branch events was counted;and the execution time of java method was calculated.With this information,a hot spots decision algorithm was proposed,and a hotspot prediction algorithm with history information was presented.The algorithm was simulated based on J2ME CDC on a MIPS platform,and the benchmark results using CaffeineMark show that the overall performance was improved by 2%,and the loo Ptest performance was increased by 6%.