> - In your example, you say that 3M is still large? Why do you think
> the actual changes smaller? Why don't you ship the delta of each files
> instead of each complete file?
> 3M is still a pretty long download for a modem, but it's not horrible.
> We could ship a delta of the individual classes (binary diff the
> actual class files too) but we chose not to do that because we
> decided that 3M was small enough and the extra complexity (and
> possibility of bugs) was not worth the potential gain for our use-case.
As a point of interest you could still use pack200
on the JAR that contains only the changed/added files. That would
may well cut the required space in half.