From mboxrd@z Thu Jan 1 00:00:00 1970 From: Gregory Leblanc Subject: Re: some IDE trays are bad (was Re: SW-RAID 1 and kernel 2.4.18) Date: 03 Dec 2002 14:10:16 -0800 Sender: linux-raid-owner@vger.kernel.org Message-ID: <1038953416.1917.19.camel@peecee.linuxweasel.com> References: <984305F1827DD5118FEA0000D1105F0A01BDC029@HW_MAIL> <20021203160141.GA7990@apartia.org> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20021203160141.GA7990@apartia.org> To: "'linux-raid@vger.kernel.org'" List-Id: linux-raid.ids On Tue, 2002-12-03 at 08:01, Louis-David Mitterrand wrote: > On Tue, Nov 26, 2002 at 02:56:41PM +0100, SCHEP. - Schepke, Arnt wrote: > > The system seems to work correctly. But i get some errors. In the file > > messages i found: > > > > linux kernel: hdc: dma_intr: status=0x51 { DriveReady SeekComplete Error } > > linux kernel: hdc: dma_intr: error=0x84 { DriveStatusError BadCRC } > > I've just had the same problem. My setup: 4 Maxtor 160G connected to a > Promise 133TX2 card, one of them was in an Icy Dock removable rack, the > three remaining were directly connected to the IDE cable. > > The tray-connected disk would always give us the BadCRC error upon > resynching with the raid5 array. That kind of error is typical of bad > cabling. > > It seems the additionnal connectors and cable-length in these IDE trays > is too much to bear for picky/sensitive ATA133 drives. In any case the > BadCRC disappeared the moment the drive was directly connected the IDE > ribbon. It's probably far more the quality of the connectors and such than the length of the cables. > hdparm -t /dev/md1 gives me over 100MB/s on IDE raid5. Incredible! Over > twice what a top-of-the-line 15k scsi drive gives me. Are you saying that a RAID 5 array gives you better performance than a single disk? Make sense to me. I'd be far more interested in a real benchmark, rather than hdparm. Across given runs on a single disk, I've found it to be reasonably reliable, but from one disk to another, not necessarily. Greg -- Gregory Leblanc