Experimental evaluation of design tradeoff in specialized virtual machine for multimedia traffic in active networks

In active network environments, the data packets can carry active programs to enable specialized processing on them. We quantify the effectiveness of general capsule programs vs. specialized processing functions for multimedia data through four experiments. These experiments deal with the comparison of Java vs. C implementation of a MPEG video decoder, the identification of the time-consuming modules in a Java MPEG video decoder, the effectiveness of combining Java methods and native methods, and the size of the bytecode for each module in a Java MPEG video decoder. We found that a Java MPEG decoder can be 2.6 times to 10 times slower than an equivalent C implementation. We identified the Huffmann decoding module as the most time-consuming module. We also found that the Native Method Interface (NMI) is complex and not efficient enough for use in active routers and the size of the bytecodes for most of the modules is too big to fit into a single packet even after compression. We draw certain conclusions about the trade-off between the general programming model and the specialized functions provided by the router for the active capsules.