Problems with db-1.85's hash indexing scheme.

Peter Whittaker (pww@entrust.com)
Mon, 12 Aug 1996 08:58:43 -0400

>From what I've seen on this list, many slapd users of db 1.85 use the
btree index scheme, and use it quite successfully. How about the hash
index scheme? We've seen problems with how page inserts are done, and
have been attempting to correct them, using Seltzer and Yizgit's paper
as a guide, with only limited success. The numbers quoted below for the
btree scheme seem beyond our grasp with hash.

Have any of you encountered - and hopefully fixed - these problems?

pww

Peter Whittaker [~~~~~~~~~~~~~~~~~~~~~~~~~~] X.500 Specialist
pww@entrust.com [ http://www.entrust.com ] Nortel Secure
Networks
Ph: +1 613 765 2064 [ ] P.O. Box 3511,
Station C
FAX:+1 613 765 3520 [__________________________] Ottawa, Canada, K1Y
4H7

>----------
>From: jrd%bu.edu@bnr400[SMTP:jrd%bu.edu@bnr400]
>Sent: Friday, August 09, 1996 1:17 PM
>To: bshafer%du.edu@bnr400
>Cc: ldap%umich.edu@bnr400
>Subject: Re: slapd - which backend to use?
>
>On Fri, 9 Aug 1996, Bob Shafer wrote:
>> My question is this, is one better than the other? Has anyone looked
>>at the
>> relative performance, etc. ?
>
>I think that the index attributes are the most important
>determinant of performance for this application.
>
>I use BSD 4.4 DB, with B-tree indices on the key array.
>My operating platform is Solaris 2.5, (Ross 166 MHZ Sparc, 128 MB core)
>
>51,000 entries - using attributes similar to the 'umichPerson'
>objectclass. Queries are blazingly fast (under a second, using
>the last name). Inserts and updates average about 1.5 seconds
>per entry.
>
>When I get time someday, I'd like to build a test-suite to
>query the database randomly with concurrent connections.
>I also have data generated by our existing Ph/CSO driven mail-server
>(phquery) to hot-test the LDAP server with real-world data.
>
>If anyone is interested, I have built tools that will allow you
>to easily migrate/synchronize your existing CSO database to SLAPD.
>It uses mSQL as a intermediate stage to warehouse the data and
>convert it into LDIF.
>
>Anyways, My database looks like this:
>
>------------------------------------------------------------------
>index cn
>index sn eq,sub,approx
>index default none
>------------------------------------------------------------------
>
>+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
>+-+-+
>Jim Doyle Boston University Information
>Technology
>Systems Analyst/Programmer Distributed Systems Group tel:
>617-353-8248
>"UNIX for a stronger America" email: jrd@bu.edu fax:
>617-353-6260
>-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-++--+-+-+-+
>-+-+-
>
>
>