Ben wrote:
>On Sun, Jul 13, 2003 at 01:08:01AM -0700, Bill Volk wrote:
>>
>> The JVM is quite slow as it is ... this would be even worse.
>Recall that very small sections of a program's code - normally
>low-level routines - account for most of its running time. As long as
>those all fit into the JAR then an interpreter could be used for the
>high-level code that doesn't account for much time. Consider the code
>used for Infocom games (of course, those aren't time-critical).
>Alternatively the demand-loaded "code" could just be maps or other
>data tables, as dc suggested.
Our experience on the speed issues is based on trying to port our image
decompression system to MIDP 1.0. We had a 5:1 image compression system
and figured that getting this on MIDP 1.0 would speed up transmission
time.
We wound up creating a PNG encoder so we could deal with raw pixels.
Wound up being a zero-sum game ... the speed we gained with downloads
was eaten up by the J2ME code running the decompression and PNG encode.
This was a compressor that could decode a 320 x 240 image in 1/4 second
on a 32MHz 386 ... so the J2ME interpreter is pretty pokey.
By the way, I was the technical director and did the game engine for
"The Return to Zork" in 1993.
I have been contacted by a vendor who claims to have the code for
downloading code objects and executing them on MIDP 1.0 ... we will see.
Bill Volk
Received on Sun Jul 13 20:33:58 2003