|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
Subject: Re: CDB tables ?
From: Mark W. Eichin (eichin
thok.org)Date: Mon Sep 11 2000 - 12:02:36 CDT
- Next message: Lorrie Wood: "Re: sendmail traces"
- Previous message: Michael Ju. Tokarev: "Re: Bypassing .forward, how?"
- Next in thread: Brad Knowles: "Re: CDB tables ?"
- Maybe reply: Mark W. Eichin: "Re: CDB tables ?"
- Reply: Brad Knowles: "Re: CDB tables ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
> Myself, I have seen Berkeley db perform exceptionally well with
> very large inputs (we stress-tested it with millions of alias entries
> when I was working on the next-generation GNN mail system, just
Be very careful about versions and modes here; I stress tested
Berkeley db 1.85++ in Hash mode in the millions of entries range using
kerberos principals, and ended up finding (and fixing) some *major*
lose-the-whole-database bugs. We ended up switching to BTree mode
because it was far better code, and ended up being *significantly*
faster at the ranges we care about (I believe the ratio of key size to
data size is relevant here, with hash collision being a secondary
factor.)
I'm pretty sure the fixes got into db v2 but I never had an
opportunity (or motivation, really) to try the tests again...
- Next message: Lorrie Wood: "Re: sendmail traces"
- Previous message: Michael Ju. Tokarev: "Re: Bypassing .forward, how?"
- Next in thread: Brad Knowles: "Re: CDB tables ?"
- Maybe reply: Mark W. Eichin: "Re: CDB tables ?"
- Reply: Brad Knowles: "Re: CDB tables ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]