[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: (ITS#7743) bdb_idl_intersection() seems to expand the search candidates unnecessarily

Dear Howard,

Thank you for your prompt reply.

 >This is certainly no bug in idl_intersection(), it is an
 >optimizatio that makes subsequent intersection operations faster.
 >As such, your suggested fix is wrong.

I understood the line 1094-1099 part of idl.c was implemented
to optimize subsequent intersection oprations.
On the other hands, in this case,  bdb_idl_intersection(C,D)
(as I described before) returns the range-type candidate set
despite its having only two elements.
Then, by subsequent bdb_idl_union() with A (list-type: the number of 
elements=296, minID=296,maxID=858689), the candidate set becomes
very huge.
#bdb_idl_union() seems to change the candidate set into "range-type"
#if one of the two sets is a "range-type".

The impact of removing the optimization of bdb_idl_intersection() is
only that subsequent bdb_idl_intersection() oprations become slower
in certain cases?
If the aforementioned optimization becomes effective when the number of
the element is more than certain threshold number(ex.1024),
I would like to change bdb_idl_intersection() like that:

  1091          if ( BDB_IDL_IS_RANGE( b ) 

  1092                  && BDB_IDL_RANGE_FIRST( b ) <= 

  1093                  && BDB_IDL_RANGE_LAST( b ) >= 

  1094                  if (idmax - idmin + 1 == a[0] && a[0] > 1024) 

  1095                  {
  1096                          a[0] = NOID; 

  1097                          a[1] = idmin; 

  1098                          a[2] = idmax; 

  1099                  }
  1100                  goto done; 

  1101          }

The issue occurs in the environment of our user and makes some trouble
because of the slow response of the ldapsearch.
So, I would like to find some fix of the issue.

 >If you want to suggest a fix, you should look at idl_union() instead.

Thank you for your advice.
I'll also try to check bdb_idl_union().

If you have other good idea to fix the issue, could you let me know?

Best Regards,
Yoshinori Nishino
NEC Soft, Ltd.
E-MAIL: nishino-yoshinori@mxc.nes.nec.co.jp