From owner-freebsd-hackers  Mon May 15 09:56:58 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id JAA07134
          for hackers-outgoing; Mon, 15 May 1995 09:56:58 -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 JAA07127
          for <hackers@FreeBSD.org>; Mon, 15 May 1995 09:56:54 -0700
Received: (from bde@localhost) by godzilla.zeta.org.au (8.6.9/8.6.9) id CAA04994; Tue, 16 May 1995 02:53:15 +1000
Date: Tue, 16 May 1995 02:53:15 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199505151653.CAA04994@godzilla.zeta.org.au>
To: hackers@FreeBSD.org, wosch@cs.tu-berlin.de
Subject: Re: slow floppy
Sender: hackers-owner@FreeBSD.org
Precedence: bulk

>My floppy is very very very slow.

>copy 1.2 MB:
>1) mcopy 50 sec
>2) cat 165 sec
>3) dd 165 sec
>4) mount -t msdos 585 sec (!!!)

>4) mean that I need 1 1/2 h for installing bindist! A poor null modem
>is faster.

cat and `dd bs=18k' on the raw device should be faster.  The default
block size for /dev/fd0 is stupid (2k) while the default block size
for /dev/rfd0 is better (16k) although it still has nothing to do
with a hardware.

The good performance for mcopy is probably because it uses the raw
device with a large block size and the especially bad performance
for msdosfs is probably because it uses the tiny file system block
size of 512 bytes.  msdosfs apparently reads only one block at a
time and your system is apparently just slow enough that reading
adjacent blocks in separate i/o's requires waiting a full disk
revolution between the blocks.  Thus you get a speed of about 1/5
second per block and it takes at least 2880 / 5 = 576 seconds to
read the whole disk.

To fix this, msdosfs needs to be improved.  It doesn't call the vfs
clustering routines.  It needs to call them or do its own clustering.
(The original version of it should have done its own clustering since
vfs clustering didn't exist then.  ufs didn't suffer so much from
the lack of clustering because of its larger block size.)

Bruce