From owner-freebsd-current  Sat Apr  8 03:28:53 1995
Return-Path: current-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id DAA09662
          for current-outgoing; Sat, 8 Apr 1995 03:28:53 -0700
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.34])
          by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id DAA09650
          for <freebsd-current@FreeBSD.org>; Sat, 8 Apr 1995 03:27:36 -0700
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id UAA24575; Sat, 8 Apr 1995 20:21:11 +1000
Date: Sat, 8 Apr 1995 20:21:11 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199504081021.UAA24575@godzilla.zeta.org.au>
To: freebsd-current@FreeBSD.org, taob@gate.sinica.edu.tw
Subject: Re: Disk performance
Sender: current-owner@FreeBSD.org
Precedence: bulk

>    My feeling is that it should have been lower and not anywhere
>close to 100% usage.  Sending out 366 I/O requests to a SCSI device
>and waiting for them to return did not seem to warrant a 50% busy
>state with a 100-MHz processor on a 33-MHz bus.  I gather this is
>where IDE drives fare much worse?

Actually only 366/8 i/o requests are sent to SCSI devices.  Iozone
does huge sequential i/o's on which clustering works perfectly,
so file data is always read and written 64K at a time (not 8K for
a file system with a block size of 8K).  Normal file accesses aren't
as sequential as for iozone, so clustering doesn't work so well.
Normal file accesses are often 8 times as slow as for iozone for
this and other reasons (seeking...) :-(.

For IDE, drives, sending out 366/8 I/O requests is much faster, but
"waiting" for them to return actually requires handling up to
366*16 interrupts (one for each sector) and copying 512 bytes or
more per interrupt.  Interrupt overhead is about 5usec/interrupt
on a P90 and copying overhead is about 155usec/sector for the old
IDE interface on all systems.

Bruce