>>>> "The SPARC architecture will not go way. It will remain the heart of SUN's server product. The future sparc processors will have architectural features that will enable Java byecodes to run very fast.
Has Sun announced this, or are you speculating? <<<<
Sun has talked about their future strategy in a
barage of announcments. The first step is to build
Java (JVM ... etc.) right into the product. The next
nature course of actions is to build speed up features
into Sparc, like what Intel is doing with Merced.
>>>>>
In any case, if they do this, and it significantly speeds-up running Java byte code, it is essentially what I am asking for. Whether or not it is a complete implementation of Java byte code is unimportant - the result is.
"Java chip are designed to be simple high performance processor that goes into client devices ... like PDAs, mobile phones, network computers, or more powerful multimedia client workstations. The key to Java chips is that the simplicity in the architecture (even simpler than RISC) allow it to be a low cost, solution for business and home market."
So, Java chips are only suitable for certain applications? Why is that? If the language is suitable for all applications, why aren't the chips?
<<<<
Java chip theoretically can be suitable for all kinds
of applications, but I think for marketing purposes these chips
will go into the networking appliance first. There are plenty
of 32/64 bit RISC processor for desktop and servers. Java
architecture at its present form may not be the best solution.
You have to understand, we are talking about chips that
takes about 60 people 1 man year to design, build, test
and ship. It is a very simple machine more of a simple
microprocessor/embedded processor. And not a Sparc or
X86 chip which, take 200-400 persons about 3-5 years to build.
Besides those chips will have features to speed up Java.
They will run Java Bytecode very fast even faster than some
Java chips, but at a *much* higher cost.
>>>>>
Can you explain how they are "simplier than RISC"? I have not seen anybody make
this claim to date.
<<<<<
Java Architecture is based on a stack machine, or
a zero address machine. I won't bored you with details, but
The Java chip organization looks more like the inside of a
HP calculator than does a RISC microprocessor. Otho, the Java
chip will have the ability for superscalar, branch predictions ...
etc.
>>>>
"The Dual instruction set strategy will delay processor development time and will
increase cost exponentially. A better way to maintain legacy is to simply include
Java chip as a add on co-processor. When the Java chip comes out, there will be a
add on PCI card that you can buy for about $100 WinTel PCs that will allow you to
run Java applications much faster at very low cost. I think this will happen sometimes
next year."
Historically, this has never been a successful approach. However, perhaps it is
"different this time".
<<<<
We'll see. Since the Java chips are so cheap, who
knows how they will manifest themselves? Maybe in a couple
of years, a standord x86 mb spec will include a Java coprocessor.
>>>>
I think it's likely that having to install an accessory card in a precious PCI slot (they
are still precious, though this is slowly changing are more PCI slots and fewer ISA
slots go into new computers) will be a serious disincentive to running Java
applications. It's more likely that users will just try to run them without the card,
conclude that Java is too slow, and go back to their old applications.
<<<<
That is a concern for home users and PC enthusias.
However, for business that have old 480/Pentium. This will be
a geat way for them to reclaim those old computers and turn
them into viable business terminals (Or ad hoc Network Computers).
Maybe package the Java card with networking support like ethernet
card and/or JavaOS or something. This could potentially be a
big market.
- Chung |
