Dear Mian,

The bitvector class performs bitwise logical operations as much as 
possible in compressed mode, however, there are cases where producing 
a uncompressed result is faster, then we produce uncompressed results. 
  Our VLDB 2004 paper deals with this kind of choices in detail 
<http://lbl.gov/%7Ekwu/ps/LBNL-54673.html>.  There are additional 
details on this issue at <http://lbl.gov/%7Ekwu/ps/LBNL-49627.html> 
and <http://lbl.gov/%7Ekwu/ps/PUB-3161.html>.

There are a number of functions in ibis::index class to sum (OR) a 
number of bitmaps together, e.g., sumBits and addBits.  These 
functions are based on the discussion from VLDB 2004 paper.  You might 
consider using them instead of your own.

Now that we are on the issue of whether you can use part of the code 
without the rest, I see two related issues.  On the patent right, as 
long as what you are doing is for research purpose only, you will be 
permitted to use the WAH compression implemented in ibis::bitvector 
class (under the fair use exception).  On the software side, you are 
technically still using FastBit code, therefore, must respect FastBit 
license (LGPL).  By the way, I am not a lawyer and am not authorized 
to speak on behalf of my employer (who holds the patent right and owns 
the software) on this matter, therefore my answer is purely a personal 
opinion.

John


On 10/16/2009 2:41 AM, Mian Lu wrote:
> Hi John:
> 
> I am studying your fastbit code recently. I have tried the AND operations
> on two bitvector classes, but found that the direct result vector looks
> like not compressed (I observed the output result through the print
> function). Only after the function compress() is executed, then it is
> compressed. Am I right? I want to ask whether this way is more efficient
> than the direct compressed result is produced? Moreover, if I want to
> perform multiple AND on multiple bitvector classes, then the compress()
> should be called after each operation?
> 
> Since I also have tried to implement the WAH algorithm by myself, but not
> so efficient compared with your fastbit. Therefore I want to adopt some
> codes from your fastbit in my research project (I hope it is okay).
> Thanks.
> 
> Regards,
> Mian
> 
_______________________________________________
FastBit-users mailing list
[email protected]
https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users

Reply via email to