Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!usc!apple!motcsd!motsj1!mcdchg!att!cbnewse!curtin
From: curtin@cbnewse.ATT.COM (C.S.Curtin)
Newsgroups: comp.databases
Subject: clipper internals
Keywords: clipper dbase index
Message-ID: <13221@cbnewse.ATT.COM>
Date: 22 Feb 90 22:39:17 GMT
Distribution: na
Organization: AT&T Bell Laboratories, Naperville, IL.
Lines: 60


       
       this posting is for a friend without net access. I will forward
       e-mail and redirect flames > /dev/null.


       dBase/Clipper gurus

       I am doing an application that is having speed problems when
       many  records  are  entered  into  the  database.  The  time
       seek()ing records using the index do not seem to be  linear.
       Sure  with  additional records the seek time should increase
       but the factor seems to be  x**2.   This  generates  several
       questions  for the data base internals gurus to answer.  Are
       there any good books on  the  internals  of  pc  bases  data
       bases? (mainly clipper/dbase)

         1.  How are the index files  structured?   The  data  base
             will  have  approximately 20,000 records and will have
             several indexes active at a time. currently  the  size
             of  the  index  file  is  250K bytes. How is the index
             internally accessed by clipper, since the index cannot
             be  kept  in  core memory?  Does it require clipper to
             access the index file on disk first?

               a.  does anyone have any  hints  on  increasing  the
                   speed of the index access?

               b.  Does index key data type have that  much  to  do
                   with  it?   I  am  assuming  that it just does a
                   string compare between the requested key and the
                   key in the index file.

               c.  What does it cost to index on various data types
                   (ie.   string+int)  or  should  all  indexes  be
                   performed on just string "types" ?  key

         2.  How is  the  skip  command  performed?  What  actually
             happens  when  a skip command is given?  Does the file
             pointer just increment X bytes (whatever the width  of
             the  record)  number  of  bytes, or is the record read
             into the buffer and flushed by clipper?  It appears as
             though the record is read into the buffer and the file
             pointer not just incremented


                     is it?
                             lseek(dbf_fd, (num_skips)*sizeof(record), SEEK_CUR);

                     or?

                             while (num_skips--)
                                     read(dbf_fd, rec, sizeof(struct rec);


       Any assistance on these issues  or  others  related  to  the
       internals  of  clipper/dbase  would  be  appreciated.  I  am
       concerned about the speed of the data access  and  currently
       cannot  see how a system with many records can be maintained
       using either of these databases.