Automatic Generation of Optimized Code for Marshalling Routines

We describe a new approach to the automatic generation of marshalling code which results in code that is fast and compact at the same time. The key insight behind our work is that certain types in an interface specification occur more often than others at run-time. We exploit this locality to solve a particularily difficult optimization problem in stub generators, namely the trade-off between compact but slow interpreted marshalling routines and fast but large compiled marshalling routines. We resolve this trade-off by using a hybrid between the two techniques. In a simulation study with ASN.1/BER we find that the hybrid approach can lead to fast and compact presentation conversion routines: the results indicate that for the X.400 e-mail protocol 50 % of the speed benefit of the compiled marshalling technique can be achieved by investing only 20 % of its code size overhead.