|
Sal, when computer people talk about bytecodes and bytecode machines, they almost always mean
something designed to be interpreted in software. Wirth's original Pascal implementation was based on a bytecode virtual machine.
Eventually, someone tried to implement it in hardware, but software implementations of P-code, as it was called, were more common and successful. Smalltalk was also based on a bytecode architecture; I don't recall anyone trying to implement it in hardware.
......................................................................
The old rule of thumb was that software emulation was about 10x as expensive as hardware implementation.
This was usually used in the compilation context though. That is, a Pascal bytecode interpreter would be expected to be about 10% of the speed of Pascal compiled to run on native hardware.
Language oriented hardware has been attempted at various times, but in general compilation to native code has proven more effective.
......................................................................
JVM is as much a specification technique as an implementation framework. I don't think Sun has any illusion about making their Java chips as fast as the latest Pentium hardware; good luck to them if they think they can.
If they can make cost effective chips, that run Java as fast as a bytecode interpreter on the lastest, fastest PC, but at much lower hardware cost, I think they'll be happy.
NC's aren't supposed to do heavy computation. And Java bytecodes compiled to run on native hardware will probably always be faster. It's not necessary for NC's to have the fastest possible hardware, only that they be fast enough to provide a good user interface. Cheers, Dan. |
