Two Topics On Dynamic Binary Translation And Optimization

Dynamic binary translation and dynamic optimization are new ways to migrate applications and to improve system performance. Researches are carried out in this field along with many influential systems built. In dynamic binary translation and dynamic optimization, works that used to be done at compile time are delayed to runtime, which demands more on the control of overhead. Despite of the overhead limitation, we are provided with many optimization opportunities at runtime that are not available at static compile time. Thereby dynamic binary translation and dynamic optimization should take different methods from static compilation in that considering more feedback than analysis, demanding more efficiency than optimum. The thesis mainly comprises two parts:In the first part, issues about translation efficiency and quality are discussed. We figure out frequent translation patterns according to statistical analysis and encode the patterns into the translator as meta-information. Once the patterns are matched during translation, source object code segments are easily mapped to target object code. The pattern based translation method not only improves translation quality, but also reduces runtime overhead.In the second part, we compare two hot trace prediction schemes. first we propose an edge profile based hot trace prediction scheme called EPB, then introduce an famous hot trace prediction method Dynamo, both are implemented in our system. Experiments show that EPB produces hot trace code with more locality than Dynamo, and behaves better in runtime performance.