Well, I think that this might be a problem on mirror break if the device names were reassigned to a different drive on reboot!

I had this same problem under windows when a RAID-1 mirror would break. And if you are using the machine in any type of production environment this really must be addressed so you don't encounter a service outage. If you catch the break and you have hotswap ability then maybe you are ok but most don't have hotswap on IDE so what happens is whenever the server next boots the device names are all wrong due to the broken mirror and the server behaves unpredictably and in some instances has actually caused damage on one of my systems.

Regards,
Gerry Reno

Thats true but you can specify your root partition inside lilo or grub and once you boot, you can modify /etc/fstab and reboot again to a use your fully working drive in an hd crash.

The 2nd posibility is create a new lilo / grub entry with the new root partition and use rdev with your kernel binary to change the default fstab. If you do this you only need to select this kernel on an array crash. The problem of this method is that you can select this kernel accidentaly and do changes in only 1 of the drives.

I suggest the 1st method

What I'm really looking for is a way to configure the system such that if a RAID-1 array does break and the server reboots that the system will automatically come up in a working configuration without any need for manual intervention so that no service outage occurs. My preferred solution would be for Linux to allow me to manually designated which device names apply to which devices and for those devices the system does not try to reassign device names. This would mean that on a RAID-1 system you might have only hda and hdc showing while the array was mirrored but would have hda,hdb and hdc,hdd when the array was not mirrored. If this can't be done then some other automatic way that on reboot the system would determine which of four possible device configurations it should be using (2 arrays with 4 possible states). Can this be done?

Regards,
Gerry Reno

To maybe clarify what I do under windows:

This all works fine except if primaries fail which is why I always make oldest drives as secondaries as they are most likely to fail first and when I replace them I make the new drive the primary.

I was hoping to do something similar under Linux.

Regards,
Gerry Reno

The sistem cant do this automatically because the raid bios will ask you if you want to stop using de array when the system reboots.

Yes, but if a sysadmin just says OK to not using array and lets system boot then you have problem. Probably they should remove bad drive before reboot if they don't have spare. That would be better.
I'm looking to make this as foolproof as possible because I have run into this several times already. Maybe there is no way to do this easily.

Regards,
Gerry Reno

I think that the better way can be a boot floopy (with a kernel that watches for a diferent fstab file and root partition, you dont need more) or replace the bad drive rebuilding the array in the bios,...

That 2 are really fast an secure methods...

I agree. Have to make sure we always have spares on hand. The problem with waiting to rebuild the array in BIOS is the time that it takes to do it. Does HPT have any rebuild-on-the-fly capability? So you can bring up the server and it will rebuild the array while it is running?

Regards,
Gerry Reno

I'm not sure about this but i think that it's posible to do this if you want to mirror an unmounted drive.

Hi Shark, I was curious what kind of performance you were getting from
the HPT370 Raid setup. I got mine running a couple of weeks ago...
but I question if it was worth the trouble....
Now that RH see's the raid as a scsi device hdparm is pretty useless..
although I can do a hdparm -t /dev/sda and get my transfer rate...very poor.
average was about 26 meg per second..Running a pair of 7200 rpm-ATA 100 drives in a raid 0 I thought I would see much higher numbers..
I've also been working on getting my ATI Radeon 8500 working decent...
hasn't happened yet..@#_)@(#_)!(_# :))
Over all the machine is not as fast as I would expect..I hate to say it but it ran better under Win2k and god help me even......ahhhhhh XP....

Dual P3 1000's
768 megOram...

My array works at 60MB/s with the scsi emulation driver and faster with the gpl driver inside the kernel, ... but really cant do hdparm -Tt with the kernel driver :P

#hdparm -tT /dev/ataraid/d0
/dev/ataraid/d0 not supported by hdparm

If you try the gpl driver (the driver included inside 2.4.19-ac4 works, ... i tried lots of versions, ... all failed at boot time). if you do:

# hdparm -X 69 /dev/hdg ; hdparm -X 69 /dev/hde

/dev/hdg:
setting xfermode to 69 (UltraDMA mode5)

/dev/hde:
setting xfermode to 69 (UltraDMA mode5)

With the gpl driver you can obtain better transfer rates.

My separate samsung 80 GB / ATA 100 transfers 40 MB/s with this configuration.

By other hand, ... if you have Barracuda IV drives you will get really low transfer rates. The solution is call seagate to exchange your drives. (They have drives with a better firmware designed for raid).

Hmmm 3 weeks to get the HighPoint driver working and now to start over..ahhhhhhh..falls on floor...

So your getting 60 megs transfer with the emulation? Why is mine so slow? My drives a pair of Quantum Fireball Kx's ATA 100 7200 rpms...I would figure I would do better then 26.

Well I guess I'm off to locate the 2.4.19-ac4 kernel.
Tanks

upgrade from 2.4.2-2 to 2.4.9-31enterprise on 4SDA+ Epox board
 

First off: I am truly a linux newbie, and therefore THANK GOD for this thread, as it has gotten me so much further than I would have imagined ever.

I installed RH7.1 on the EPOX 4SDA+ board, which has the same HPT372 controller as the board that originated this thread. There are 2 WD8000JB drives connected to the HPTcontroller, and these are the only drives in the computer. They have been setup as RAID 1, since I really don't want to lose any work.

I found the source code for the 9-31enterprise kernel somewhere, I copy the configs/2.4.9-31[enterprise] config file to the .config in the source root directory, make menuconfig, disable the hpt controllers in ATA/IDE/MFM/RLL, edit the drivers/ide/ide-pci.c to drop the DEVID_HPT34X and DEVID_HPT366 lines (can not find the other ones mentioned in the post 32), downloaded the lastest opensh*t source codes from highpoint-tech.com, (v1.31, in which I think they fixed the gcc3 issue ??), since my gcc is at 2.96 something I am fine here, I compile the HPT driver succesfully.

make dep clean
at this point I compile the HPT driver
then continue to
make bzImage modules modules_install install
(I have found that if I compile the bzImage before the modules, I don't get error messages during the compile of the modules)

at this point I copy the hpt37x2.o driver to
/lib/modules/2.4.9-31custom/kernel/driver/scsi

then run depmod -a (after I cd'd to the above directory)

then run the mkinitrd command as described in post 49 and others:

/sbin/mkinitrd -f --preload scsi_mod --preload sd_mod --preload ext3 --preload hpt37x2 --with hptraid /boot/initrd-2.4.9-31-custom.img 2.4.9-31custom

I created an additional entry in my /etc/lilo.conf file to address my new kernel:

install=/boot/boot.b
prompt
timeout=50
message=/boot/message
linear
default=linraid

[..]

image=/boot/vmlinuz-2.4.9-31custom
label=lincustom
initrd=/boot/initrd-2.4.9custom.img
read-only
root=/dev/sda5

This works, but at the welcome message for Redhat linux, the bootup procedure stalls ...
right after the line 'Type I for interactive setup'

I figured that some of the initial parameters of the mkinitrd may hose up my system, so I recreated the img file as follows:

/sbin/mkinitrd -f --preload scsi_mod --preload sd_mod --preload hpt37x2 /boot/initrd-2.4.9-31-custom.img 2.4.9-31custom

But the startup stalls at the same point. _SIGH_

The reason I want the 2.4.9-31 enterprise kernel, is because this is the last version still supported by the application I load on this machine.

Any ideas anybody?

I am lost at this point.
I have just recompiled the kernel AGAIN

Thanks for any support

The 'safety' of RAID 1 setup seems slim ..
 

Hello all again:

From my former post you can see the woes I am having in getting the RAID1 system running. When I read all the posts, I couldn't help but notice that the failover was not nearly as automatic as I had hoped.

THerefore, I have the following question:

Should I setup the two identical drives as two separate harddrives, and fully run everything on one harddrive, and the have the second be a spare, that is connected, and is used to backup the entire drive periodically (on demand).

I am thinking about this, because it is sheer dumb luck (+ an attribute to the linux operating system) that I have a working system still. The kernel upgrades are on both disks simultaneously, so if I would have hosed up the system, the RAID1 would not have helped anything.

What command/application would I use (dd ?). Could you schedule this? how?

If I setup the two drives in JBOD mode (and NOT on the HPT372 controller, thank you very much), could you make a striped linraid volume for the swap partition (to have at least some gain of the two drives speed wise) or would this make it a very fragile system?

Questions quesions

Some day I might be able to answer some one else's questions on linux.

Until then, thanks in advance for all your (moral) support