Path: utzoo!utgpu!watmath!clyde!att!osu-cis!killer!ames!ncar!boulder!sunybcs!rutgers!bpa!cbmvax!uunet!mcmi!denny
From: denny@mcmi.UUCP (Dennis Page)
Newsgroups: comp.sys.att,u3b.tech
Subject: Re: Question about 2kb filesystems
Message-ID: <600@mcmi.UUCP>
Date: 4 Jan 89 22:50:34 GMT
References: <978@vsi.COM>
Reply-To: denny@mcmi.UUCP (Dennis Page)
Distribution: comp
Organization: MCMI, Omaha, NE
Lines: 23

In article <978@vsi.COM> friedl@vsi.COM (Stephen J. Friedl) writes:
>     A customer of mine has a 3B2/600 running System V Release
>3.2.1, with an Informix-SQL application (ver 2.10.00B).  We just
>converted the filesystem with the actual *large* database files
>to be a 2kb filesystem, and the customer notices that everything
>runs slower.

Did you test the rotation gap when you changed logical block sizes?  What
is the time difference between moving 1k & 2k blocks?  Did they change mkfs
to use *logical* gap size instead of physical (512) gap size?

Did you check to make sure that the data area still begins on a cylinder
boundary?  (Twice the logical block size means twice the boundary size for
superblock & inode tables.)

So many questions!  :-)

Btw: Be wary of the possibility of mkfs using logical blocks for gapping
and fsck using physical blocks.  This nasty screwup, especially if you do
fsck -s by habit.  I don't know if AT&T 3.2 has this disease (I really
doubt it -- they don't have it in 3.1.1) , but Unisys 3.x sure does.
-- 
Good health is merely the slowest rate at which one can die.