The experiments in the previous sections all use 8-byte messages. To see how varying the message size affects the savings in time and space, we have performed the same set of experiments with larger messages (64 bytes). The measurements follow a similar trend, but the percentage reduction in execution time is less than when using 8-byte messages. This is because the execution overhead of the actual message buffer read/write operation, which cannot be reduced, becomes a more dominant part of the total execution overheads. The percentage reductions in space overheads are the same, or slightly better than for the 8-byte message case, since the constant overheads of some of the control variables are less apparent. Due to the substantially similar results, the 64-byte message measurements are not presented here.
We have also conducted experiments while varying the total size of the reader set. Running the previous experiments with 10 reader tasks resulted in nearly identical relative performance improvements with our transformation mechanism. Of course, with fewer readers, any complexity increase in the writer task has greater weight in the average execution time, but this is offset by the performance gains in the fast readers. Space reduction, as before, is basically linear to the percentage reduction in the number of slow readers. Again, due to their substantially similar results, the data for the 10 readers case are omitted here.