Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!lll-crg!mordor!sri-spam!wohler
From: wohler@sri-spam.istc.sri.com (Bill Wohler)
Newsgroups: comp.databases
Subject: Re: Optimization algorithms
Message-ID: <9768@sri-spam.istc.sri.com>
Date: Sun, 7-Dec-86 17:26:20 EST
Article-I.D.: sri-spam.9768
Posted: Sun Dec  7 17:26:20 1986
Date-Received: Sun, 7-Dec-86 21:34:09 EST
References: <8612010047.AA14252@BORAX.LCS.MIT.EDU>
Reply-To: wohler@sri-spam.UUCP (Bill Wohler)
Distribution: world
Organization: SRI International, Menlo Park
Lines: 39

Greg Brandon writes:

>Is there a forum specifically for the discussion of relational database
>access optimization algorithms, or is this the place for that kind of
>discussion?

  i was expecting to find more of a technical discussion when i began
  reading this column, but have not yet seen much.

  so hopefully to begin a more fruitful dialogue, i pose the
  following. i used to use "grep -i <pattern> phonebook" to retreive
  records from my personal online phone directory, but now that the
  file has begun to get quite large (> 200 records and counting), i
  have been tossing around ideas for implementing a more efficient and
  flexible method of retreival.

  however, the requirements are a little different from normal
  databases.  i'd like the generality of grep in my searches, that is,
  to be able to search for a record with only fragments of a key.  for
  example, "dan" would match the first names "Dan" and "Daniel", the
  last name "Danielson", and the company called "Danskin".  i currently
  key on first and last names, business name, and (at least now) month
  of birthday to get a quick printout of people who have birthdays
  this month. 

  dbm(3) loses since it needs fully qualified keys, and only allows
  one entry per key.  note that a name may occur more than once.

  i'm thinking of creating a multi-ring stucture, but can't figure out
  how to create an efficient way of finding an index from a partial
  key short of using a linear search.  perhaps there might be a way in
  a relational type database.

  a hash function for this would make a nice thesis (and a quick
  program). 

  well, food for thought...

							--bw