From owner-freebsd-current  Sat May 24 06:04:27 1997
Return-Path: <owner-current>
Received: (from root@localhost)
          by hub.freebsd.org (8.8.5/8.8.5) id GAA24591
          for current-outgoing; Sat, 24 May 1997 06:04:27 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id GAA24585
          for <current@FreeBSD.ORG>; Sat, 24 May 1997 06:04:23 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.5/8.6.9) id XAA06875; Sat, 24 May 1997 23:02:11 +1000
Date: Sat, 24 May 1997 23:02:11 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199705241302.XAA06875@godzilla.zeta.org.au>
To: current@FreeBSD.ORG, j@uriah.heep.sax.de
Subject: Re: cd9660 filesystem slowed down ???
Sender: owner-current@FreeBSD.ORG
X-Loop: FreeBSD.org
Precedence: bulk

>j@uriah 145% dd if=/cdrom/xxxx of=/dev/null
>13036+1 records in
>13036+1 records out
>6674533 bytes transferred in 70.361082 secs (94861 bytes/sec)

There is a stranger anomaly for ext2fs with an fs block size of 1K:

Immediately after mounting:
$ dd if=libg++.a >/dev/null
3656+1 records in
3656+1 records out
1872034 bytes transferred in 4.143877 secs (451759 bytes/sec)
This is normal - the drive is a Zip with a best case transfer speed
of about 1MB/sec.

Second run:
$ dd if=libg++.a >/dev/null
3656+1 records in
3656+1 records out
1872034 bytes transferred in 29.836025 secs (64264 bytes/sec)

Subsequent runs: about the same speed as the second run 60-70K/sec.

This has something to do with nbuf (= 0x6a5) * fs_block_size (= 1K)
being slightly less than the file size.  For smaller files, caching
gives a 20 times speed improvement instead of a 7 times speed
unimprovement.  Reading with a 4K block size doesn't help much.

Caching works even less well for writing to this file system:
$ dd if=/dev/zero of=z bs=1k count=1400
1400+0 records in
1400+0 records out
1433600 bytes transferred in 42.815710 secs (33483 bytes/sec)
$ [same with count=1000]
1024000 bytes transferred in 28.514967 secs (35911 bytes/sec)
$ [same with count=600]
614400 bytes transferred in 13.807223 secs (44498 bytes/sec)
$ [same with count=200]
204800 bytes transferred in 0.047537 secs (4308226 bytes/sec)

ext2fs with a block size of 4K seems to work fine.  So does msdosfs
with a block size of 4K.  Note that msdosfs doesn't use cluster_read().
cd9660 with doclusterread turned off is even slower than with it on,
so I assume that this is a general caching problem.

Bruce