There are probably many better ways to do this, so don't hesitate to enlighten me if you know any of them. Also, I noticed the 7.2/Roswell directories were locked up a day or two ago on ? perhaps this means a release in the near future (and perhaps a working install process for these devices?).
My original intent was to both benchmark and test the stability of the Promise drivers, Linux Software RAID, and the ataraid drivers. I had about 3 days to do this, so I figured it wouldn't be much of a problem (wrong).
I first tried to trick the RedHat install into recognizing the /dev/ataraid/d0 (link to /dev/sda, mknod /dev/sda b 114 0, etc.). When the obvious stuff failed, I decided instead of trying to find the source for the install program and figuring out if it was possible to make the install work, I'd see if I could install and boot using the Promise drivers and convert to ataraid from there (since I wanted to test the Promise drivers anyways).
* Installing with Promise drivers (I?ve seen this covered at length elsewhere, but I haven?t seen anything that covered all the problems I had).
1) Obtain relevant RedHat 7.1 drivers from: http://support.promise.com/Linux/Default.htm
2) tar xfv ftrhup_120b9.tar on a floppy (I used win2k formatted floppy (vfat) to do this).
3) On another freshly installed linux box, take ft.o (in modules.cgz from above tar file) and generate an initrd and copy the initrd file onto the above floppy. I vaguely recall reading somewhere about a second setup disk for the Promise drivers that included this initrd, but I didn't read about it until I had already figured out how to do this (so I didn't bother searching for it) - nor could I find it when I decided to look for it while writing this email.
4) Boot off CD-ROM, enter linux dd or expert at the boot prompt.
5) Put in floppy and indicate presence of driver disk - sometimes ft successfully loaded, other times not (probably due to a flaky floppy/floppy drive) - when it didn't load properly, I continued until the media selection screen (selected Local CD-ROM and hit enter) and Alt-F2'd (Ctrl-Alt-F2 if X is started) to the shell. Step 6 is what to do if it doesn?t load.
6) cd /tmp;mount fd0 drivers;cat drivers/modules.cgz | gunzip -c - | cpio -ivd;insmod 2.4.2-2BOOT/ft.o
7) Go through then install process - when it gets to the screen where it asks you if you want to create a boot disk (or to the screen where the install is finished if you didn't choose to create a boot disk), mount /tmp/fd0 /tmp/drivers again if necessary and copy the initrd into /mnt/sysimage/boot (make sure it is there and not empty before you do so).
8) vi /mnt/sysimage/etc/lilo.conf - 'initrd=/boot/initrd_file.img' (or whatever it is named) is placed in the section for the linux boot image.
9) ln -s /mnt/sysimage/boot /boot;lilo -C /mnt/sysimage/etc/lilo.conf ?v
10) Keep tweaking /mnt/sysimage/etc/lilo.conf and rerunning the above lilo command until it is successful.
11) Finish install and reboot to a login prompt (don't use a boot disk if one was created).
* Compiling and booting a kernel with ataraid drivers.
1) Place http://www.kernel.org/pub/linux/kernel/v2.4/linux-2.4.10.tar.gz and http://people.redhat.com/arjanv/pdcraid/patch-to-2.4.10 in /usr/src on your newly installed box.
2) Patch, config to taste (select all promise-related options), and make the kernel.
3) Copy image to /boot.
4) Get http://ataraid.slimyhorror.com/lilo-ataraid.tar.gz and http://ataraid.slimyhorror.com/lilo.conf. Compile and install lilo, back up your existing copy of /etc/lilo.conf and change /etc/lilo.conf using the one off slimyhorror as a guide to reflect the use of /dev/ataraid/d0p* (* == partition number - e.g. if root is /dev/sda5, change to /dev/ataraid/d0p5). Copy lilo, lilo.conf, and MAKEDEV to driver disk.
5) Change /etc/fstab to reflect changes (I removed LABEL=/ and replaced with /dev/ataraid/d0p5 as I couldn't get the ataraid drivers to boot without doing so). Run MAKEDEV script (had to change block to b to get it to work on my system).
6) Couldn't get lilo to work when the device was changed, so copied ataraid.o and pdcraid.o (2.4.2-2BOOT versions compiled on a separate fresh RedHat 7.1 install from *old* ataraid source ? which I later found out I didn?t have to do since add_gendisk and del_gendisk are apparently trivial to add) onto driver disk, booted off CD-ROM as a normal install, but typed - expert ide1=0xd000,0xd402 ide2=0xd800,0xdc02 - at the boot prompt (this will vary depending on machine).
7) insmod ataraid.o;insmod pdcraid.o (pdcraid wouldn't load if I hadn't specified ide1 & 2 above) from shell as per initial (get to and select media type first).
8) Run MAKEDEV, mkdir /boot, mount boot partition onto /boot. Run lilo from driver disk using lilo.conf from driver disk.
9) If successful, reboot should get you to a login prompt using GPL drivers.
If anyone is interested, I can provide some more detailed instructions (I wrote down all the steps I took as I was doing them along with some of the problems I encountered and their solutions) and/or disk images via email. Feel free to email me.
With bonnie++/zcav and iozone on fresh RedHat installs (2.4.2-2 for Promise and 2.4.10 for GPL), the GPL drivers beat Promise?s in just about every test. I'm not particularly skilled at performance tuning (max-readahead wasn't present on my system, so I assume it is in the ac kernels), so I didn't do any further tests. Whether or not the above configuration sufficiently isolates and/or properly tests the disk I/O subsystem especially given the different kernels is certainly debatable :), but it met my not-so-rigid requirements. Detailed output of tests available on request.
To get an idea of the results, zcav registered 35,844kB/s throughput for the Promise drivers, 42,999kB/s throughput for the GPL drivers, and a dismal 9,431kB/s for a single disk (I first tested at around 5,498kB/s, then used hdparm -c1 -A1 -m16 -d1 /dev/hde) - http://www.coker.com.au/bonnie++/experimental/bonnie++-1.92b.tgz for bonnie++/zcav, configure,make,make install, zcav /dev/ataraid/d0 (then average the speeds) and please let me know your results :).
I could only get linux to install and boot afterwards if I specified ide1 and ide2 to the kernel at boot and I don't know whether or not this would have something to do with the low throughput numbers (this is also the reason I didn't bother to test software raid at the end as I figured it couldn't have been any better than double the single disk numbers). The other IDE buses were populated, which is why I didn't move the disks there to try out software RAID (but I'd be interested in seeing the zcav test from anyone who is using it).
- Adam
_________________________________________________________________ Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp
