By: Rob Thorpe (rthorpe.delete@this.realworldtech.com), October 17, 2006 8:33 am
Room: Moderated Discussions
David Kanter (dkanter@realworldtech.com) on 10/17/06 wrote:
---------------------------
>>I thought the BCD arithmetic remark was interesting. They implied that Java and
>>other high level languages would have worse BCD arithmetic performance by a large factor. David says:-
>>"they estimate ... benchmark that BCD support could improve performance by 7x,
>>4x or 2x, compared against Java, C/C# or assembly libraries >respectively"
>
>>Normally decimal math is done in optimized libraries - often >written in assembly.
>
>Yes this was my impression. I sort of thought the 7x number which a lot of folks publicized was a red herring.
>
>>Because of this, I would expect Java to be slightly worse >than compiled languages, but not that much.
>
>That's the Java Bignum library, which I would imagine is not written in ASM.
I expect the core loops will be written in asm, probably accessed through a foreign-function call interface of some kind.
I'd be quite surprised if it wasn't. Any lisp implementation is obscure, but even some of the ones considered obscure amongst lisp users have bignum implementations with some loops optimized using assembly.
>>I also don't understand how IBM think their machine is >going to run C#, they probably mean C++.
>
>The slide really said C#. I guess IBM is of the opinion that Mono is a sufficiently
>mature package for business critical computing and financial applications on AIX.
>
>I don't know why C# was included, but I didn't think it was odd enough to ask further about.
Interesting. Also, maybe IBM have their own C# system up their sleeve.
---------------------------
>>I thought the BCD arithmetic remark was interesting. They implied that Java and
>>other high level languages would have worse BCD arithmetic performance by a large factor. David says:-
>>"they estimate ... benchmark that BCD support could improve performance by 7x,
>>4x or 2x, compared against Java, C/C# or assembly libraries >respectively"
>
>>Normally decimal math is done in optimized libraries - often >written in assembly.
>
>Yes this was my impression. I sort of thought the 7x number which a lot of folks publicized was a red herring.
>
>>Because of this, I would expect Java to be slightly worse >than compiled languages, but not that much.
>
>That's the Java Bignum library, which I would imagine is not written in ASM.
I expect the core loops will be written in asm, probably accessed through a foreign-function call interface of some kind.
I'd be quite surprised if it wasn't. Any lisp implementation is obscure, but even some of the ones considered obscure amongst lisp users have bignum implementations with some loops optimized using assembly.
>>I also don't understand how IBM think their machine is >going to run C#, they probably mean C++.
>
>The slide really said C#. I guess IBM is of the opinion that Mono is a sufficiently
>mature package for business critical computing and financial applications on AIX
>
>I don't know why C# was included, but I didn't think it was odd enough to ask further about.
Interesting. Also, maybe IBM have their own C# system up their sleeve.