Amanda-Users

holdingdisk throughput problem: Linux ide-dma

2003-08-20 11:27:19
Subject: holdingdisk throughput problem: Linux ide-dma
From: Eric Sproul <esproul AT ntelos DOT net>
To: AMANDA Users <amanda-users AT amanda DOT org>
Date: Wed, 20 Aug 2003 11:21:49 -0400
Hi all,
I am having a problem with disk I/O on a new holdingdisk I installed in
my AMANDA server.  The server had previously been exclusively SCSI, and
I had no problems.  When we broke down an old server, I snagged a 60GB
IDE drive (an IBM Deskstar 75GXP) to use as cheap extra holding disk
space-- "free" as opposed to ~$450 for a similar-sized SCSI drive.  The
problem is, the read throughput is terrible and it's causing backups to
run about twice as long, due to taper not being able to extract stuff
off the holding disk fast enough.

The AMANDA server is PC-based, with Tyan Thunder 100 (S1836) mobo and
the i440GX chipset (PIIX4 integrated IDE controller).  According to the
manual, the IDE controller should do up to ATA/33 (UDMA mode 2, I
believe).  It's running Debian testing/unstable, with kernel 2.4.21.

Anyway, I plug the drive up, setting the BIOS to auto-everything for IDE
disks.  I see the disk come up and get recognized.  However, when the
kernel detects the IDE disk, it reports no UDMA setting, leading me to
believe it came up in one of the older PIO modes.  This is confirmed
with a quick hdparm test that shows an abysmal 4.5 MB/s buffered disk
read speed.  By comparison, the same disk at UDMA(33) in a similar
machine reports 28 MB/s.

Does anyone here have experience with IDE and DMA under recent Linux
kernels?  Anyone else have a problem like this?  I figured the
integrated controller on a 440GX board would have no compatibility
problems, as it's extremely common.  Maybe I'm wrong.

Thanks,
Eric



<Prev in Thread] Current Thread [Next in Thread>