Re: [Open Babel] Java performance penalty?

[Geoff H. <ghu...@ws...>]

This is somewhat off-topic, but I thought I should at least explain my 
$0.02 from the other e-mail.

On May 22, 2004, at 4:48 AM, Miguel wrote:

> My general findings were:
>  - Overall, the performance penalty for programming
>    in Java was about 10%
>  - the worst benchmark was about 20%
>  - others were only 0-5% slower

I have a friend who works at Intel and asked me about my computational 
chemistry tasks--he wanted to know what sorts of systems we used, etc. 
I mentioned that the (then just-released) Intel compilers gave a 10-20% 
boost over GCC on calculations and I was very thankful.

I often run 1-2 week QM (DFT) calculations. Some of my manuscripts 
include 200-300 individual jobs run across clusters. When we used the 
updated compilers, I estimated I saved about 3,000 hours of CPU time 
from that 10% boost. The NCI database is 250,000 entries.

A 10% performance penalty on a 1 second task isn't very noticeable. A 
10% performance boost over 250,000 jobs -- particularly if the jobs 
might take a few minutes (e.g., MM or QM calculations) is significant, 
IMHO.

>  - I suspect that Java makes it easy for inexperienced
>    programmers to build systems with relatively poor
>    performance because of heap memory allocation/garbage
>    collection.

True, though I'll also admit that an inexperienced C++ programer who 
leaves a memory leak will produce a *much* more inefficient program 
once the computer starts to use VM.

As I said, it's just my $0.02... I *like* Java as a language. But I've 
continued to work on Open Babel in C++ because I use the program a lot. 
:-)

-Geoff