java acedb database

[David Daney]

Jean Thierry-Mieg wrote:
>>Dear java/gnu herders
>>
>>I am one of the main authors of ACEDB (http://www.acedb.org), an object 
>>oriented database system which has been used worldwide for the human 
>>genome project (http://www.aceview.org)
>>The code is in C, about 250k lines, runs on any unix ever tested, and is 
>>freely distributed under gnu-licence
>>
>>Because the code is now 15 years old, and single threaded, I consider 
>>redesigning it.
>>
>>The natural choice of language may be C++, but it seems to me that Java 
>>class hierarchy and variable scoping
>>are preferable to C++.  However, untill recently, Java could not be 
>>considered because a database kernel has to be compiled in native 
>>machine code, interpreted byte code is necessarily too slow.
>>

I might get shot for saying it here, but current commercial JVMs (like 
Sun's and IBM's) are quite good, and in many cases as good or better 
than native C++ compilers.

>>Here are my questions,
>>
>>1) considering the development of  GCJ (gcc-java) would you recemmend 
>>Java to write a large code where performance is crucial.  Critical 
>>operations are sorting, hashing, string searches. How would the speed 
>>compare to dedicated C code.
>>

If speed is the *only* criterion, it is hard to beat C.  If care is 
taken, java code can come close.

>>2) Will Java provide a performant implementation of multi threading.

Yes.

>>
>>3) Is it true that, using GCJ, Iwill be able to compile on (nearly) any 
>>unix machine (ibm,sun,mac,hp,intel....)

Current GCJ support is somewhat patchy.  I think for major platforms 
(i.e. x86/linux) GCJ is fairly complete.  For some of the more obscure 
embedded platforms it is missing support for some things (reflection) 
and some other language features (synchronization and exceptions) have 
slow implementations.

David Daney