Re: mdadm without initramfs
* martin f krafft [100816 07:33 +0200]: also sprach Boyd Stephen Smith Jr. b...@iguanasuicide.net [2010.08.16.0034 +0200]: Last I checked, it is possible to have the kernel itself start the root array, via a (series of) kernel command-line arguments. However, this doesn't use mdadm or any of its configuration files. That only works for the deprecated v0.9 superblocks and requires the partitions to be marked with type 0xFD. The whole method is deprecated. Use an initramfs. ACK Elimar -- Learned men are the cisterns of knowledge, not the fountainheads ;-) -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100816093218.ga3...@aragorn.home.lxtec.de
Re: mdadm without initramfs
Boyd Stephen Smith Jr. put forth on 8/15/2010 5:34 PM: In 20100815190053.ga4...@gandalf.home.lxtec.de, Elimar Riesebieter wrote: How do I set up mdadm to create the root array witout an initramfs? You can't. mdadm is a user-space binary that can't be compiled into the kernel images. Therefore, to run mdadm you need an initramfs, or the file system containing '/' must already be mounted by the kernel. Last I checked, it is possible to have the kernel itself start the root array, via a (series of) kernel command-line arguments. However, this doesn't use mdadm or any of its configuration files. Similarly, mdadm doesn't read the kernel command-line, so it is possible the configurations to diverge. It may be possible to have a GRUB2 hook to generate the kernel command-line arguments from the mdadm configuration, but I've not seen such a hook. In the absence of such a hook (or when you are not using GRUB2 as your bootloader), you should simply use an initramfs. This has always been one of my hangups regarding using linux mdraid (or any soft OS raid) vs hardware raid--proper and seamless handling of a raid protected boot device, including issues beyond the topic of this thread. I hate admitting it, but Microsoft's implementation of a mirrored boot/system disk is supremely simply compared to getting the same thing from Linux. Matter of fact, setting up any type/level raid in Windows is much more straightforward. And I'm not a big Windows fan, far from it. I prefer LSI and Intel raid cards. I should have said merely LSI as the Intel cards are licensed LSI cards. Hardware raid isn't as flexible as softraid as it works at the entire disk level, but boy is it so much easier to work with, as much faster. The single biggest advantage to hardware raid is that you don't have to disk with changing bios boot order or anything like that if you have to reboot while drive in your boot array is offline/down/dead. It's all automatic. -- Stan -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4c6939b1.6080...@hardwarefreak.com
Re: mdadm without initramfs
On 08/16/2010 08:14 AM, Stan Hoeppner wrote: Boyd Stephen Smith Jr. put forth on 8/15/2010 5:34 PM: In20100815190053.ga4...@gandalf.home.lxtec.de, Elimar Riesebieter wrote: How do I set up mdadm to create the root array witout an initramfs? You can't. mdadm is a user-space binary that can't be compiled into the kernel images. Therefore, to run mdadm you need an initramfs, or the file system containing '/' must already be mounted by the kernel. Last I checked, it is possible to have the kernel itself start the root array, via a (series of) kernel command-line arguments. However, this doesn't use mdadm or any of its configuration files. Similarly, mdadm doesn't read the kernel command-line, so it is possible the configurations to diverge. It may be possible to have a GRUB2 hook to generate the kernel command-line arguments from the mdadm configuration, but I've not seen such a hook. In the absence of such a hook (or when you are not using GRUB2 as your bootloader), you should simply use an initramfs. This has always been one of my hangups regarding using linux mdraid (or any soft OS raid) vs hardware raid--proper and seamless handling of a raid protected boot device, including issues beyond the topic of this thread. I hate admitting it, but Microsoft's implementation of a mirrored boot/system disk is supremely simply compared to getting the same thing from Linux. Matter of fact, setting up any type/level raid in Windows is much more straightforward. And I'm not a big Windows fan, far from it. I prefer LSI and Intel raid cards. I should have said merely LSI as the Intel cards are licensed LSI cards. Hardware raid isn't as flexible as softraid as it works at the entire disk level, but boy is it so much easier to work with, as much faster. The single biggest advantage to hardware raid is that you don't have to disk with changing bios boot order or anything like that if you have to reboot while drive in your boot array is offline/down/dead. It's all automatic. I haven't messed with windows since xp. But I do remember on xp you could not have any software raid on the same partition that you booted from. Which made it pointless... Did they change the functionality in vista or 7 or something? As far as I know, mdadm is maintained by only one man. So we better just be thankful that we have it! Sam -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4c694fea.6080...@net153.net
Re: mdadm without initramfs
also sprach Stan Hoeppner s...@hardwarefreak.com [2010.08.16.1514 +0200]: This has always been one of my hangups regarding using linux mdraid (or any soft OS raid) vs hardware raid--proper and seamless handling of a raid protected boot device, including issues beyond the topic of this thread. I hate admitting it, but Microsoft's implementation of a mirrored boot/system disk is supremely simply compared to getting the same thing from Linux. Why don't you use it then? ;) I prefer LSI and Intel raid cards. I should have said merely LSI as the Intel cards are licensed LSI cards. Hardware raid isn't as flexible as softraid as it works at the entire disk level, but boy is it so much easier to work with, … until your controller dies and you find out that the manufacturer does not support the firmware anymore and your data are lost. as much faster. Do you have research backing that up? The single biggest advantage to hardware raid is that you don't have to disk with changing bios boot order or anything like that if you have to reboot while drive in your boot array is offline/down/dead. It's all automatic. Why would you have to reboot before replacing a dead drive?? That sounds like you got your priorities wrong. -- .''`. martin f. krafft madd...@d.o Related projects: : :' : proud Debian developer http://debiansystem.info `. `'` http://people.debian.org/~madduckhttp://vcs-pkg.org `- Debian - when you have better things to do than fixing systems sailing is, after all, a kind of grace, a kind of magic. -- phil berman digital_signature_gpg.asc Description: Digital signature (see http://martin-krafft.net/gpg/)
Re: mdadm without initramfs
martin f krafft put forth on 8/16/2010 10:26 AM: I prefer LSI and Intel raid cards. I should have said merely LSI as the Intel cards are licensed LSI cards. Hardware raid isn't as flexible as softraid as it works at the entire disk level, but boy is it so much easier to work with, … until your controller dies and you find out that the manufacturer does not support the firmware anymore and your data are lost. Ever heard of spares? If not you've not been in this game long. as much faster. Do you have research backing that up? You're kidding right? If not, Google is your friend here. The single biggest advantage to hardware raid is that you don't have to disk with changing bios boot order or anything like that if you have to reboot while drive in your boot array is offline/down/dead. It's all automatic. Why would you have to reboot before replacing a dead drive?? That sounds like you got your priorities wrong. DIMM failure, extended power outage, kernel panic, colo personnel reboot the wrong box in a rack, etc, etc, etc. Sounds like you are new. -- Stan -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4c69742e.3080...@hardwarefreak.com
Re: mdadm without initramfs
also sprach Stan Hoeppner s...@hardwarefreak.com [2010.08.16.1923 +0200]: … until your controller dies and you find out that the manufacturer does not support the firmware anymore and your data are lost. Ever heard of spares? If not you've not been in this game long. I've had cases where the spares simply didn't want to work anymore, after lying around all this time, or where the surrounding hardware failed and no other hardware could actually address the spares anymore. Do you have research backing that up? You're kidding right? If not, Google is your friend here. Google has never been my friend, and likely won't ever be. So far, I have yet to meet a statistic that credibly backs up this claim, given that performance is generally bound by spindles these days. Your expensive hardware RAID card might well out-perform mdadm on a Pentium with 15000 RPM drives, but with a modern processor, I have yet to witness the kernel to be the bottleneck. DIMM failure, extended power outage, kernel panic, colo personnel reboot the wrong box in a rack, etc, etc, etc. Sounds like you are new. The chance of these happening after a degradation of an array is proportional to the admin slack in replacing the drive; and, as you said, there are spares (with a different meaning, in this case). Before you reply, consider there is a chance of two drives dying at once, or the controller going belly-up. But of course, I don't have to explain this to such a seasoned person as you. I wonder why you even bother to write in about mdadm. -- .''`. martin f. krafft madd...@d.o Related projects: : :' : proud Debian developer http://debiansystem.info `. `'` http://people.debian.org/~madduckhttp://vcs-pkg.org `- Debian - when you have better things to do than fixing systems menschen, welche rasch feuer fangen, werden schnell kalt und sind daher im ganzen unzuverlässig. - friedrich nietzsche digital_signature_gpg.asc Description: Digital signature (see http://martin-krafft.net/gpg/)
Re: mdadm without initramfs
Sam Leon put forth on 8/16/2010 9:49 AM: I haven't messed with windows since xp. But I do remember on xp you could not have any software raid on the same partition that you booted from. Which made it pointless... Did they change the functionality in vista or 7 or something? I was referring to the server products, which have no such restrictions. I believe you're thinking of a consumer oriented XP Home edition. In that case I'm surprised it offered any software RAID, period. M$ is usually very stingy in this regard. Back in the W2K era, Pro had zero RAID options. If you wanted software RAID you had to install a server version of W2K, which was 4 times the price of Pro, and in most cases more expensive than putting a good low end single channel SCSI Mylex or Intel PCI RAID card into the Pro box. At the time, $employer sold quite a few W2K Pro workstations with hardware RAID cards due to this very reason. And the customer got a performance boost for free. ;) -- Stan -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/4c698043.9050...@hardwarefreak.com
Re: mdadm without initramfs
In 4c69742e.3080...@hardwarefreak.com, Stan Hoeppner wrote: martin f krafft put forth on 8/16/2010 10:26 AM: I prefer LSI and Intel raid cards. I should have said merely LSI as the Intel cards are licensed LSI cards. Hardware raid isn't as flexible as softraid as it works at the entire disk level, but boy is it so much easier to work with, … until your controller dies and you find out that the manufacturer does not support the firmware anymore and your data are lost. Ever heard of spares? If not you've not been in this game long. Keeping the firmware versions in sync between production and spares is difficult to say the least. I've also found the kernel / mdadm maintainers a lot more responsive to issues in their stack than I have found the manufacturer of my HW RAID card when I have found issues in their firmware. as much faster. Do you have research backing that up? You're kidding right? If not, Google is your friend here. My experience is that there's no speed difference between my (at the time) expensive HW RAID system and my (at the time) expensive CPUs handling software RAID. Except in rare cases, it is more $-efficent to buy extra CPU power (on yout MB or processor shelf/drawer) that can be used by all processes than it is to buy dedicated CPU power (on your HW RAID card) that is only used for RAID tasks. The single biggest advantage to hardware raid is that you don't have to disk with changing bios boot order or anything like that if you have to reboot while drive in your boot array is offline/down/dead. It's all automatic. I've been required to enter the HW RAID BIOS interface to get a boot after a failure. I've also rebooted with failed or rebuilding arrays under the control of mdadm and did not have to deal with changing BIOS boot order. (and vice-versa, of course) I don't think it is correct to say that particular roadblock is always absent when dealing with HW RAID or always present when dealing with mdadm. -- Boyd Stephen Smith Jr. ,= ,-_-. =. b...@iguanasuicide.net ((_/)o o(\_)) ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-' http://iguanasuicide.net/\_/ signature.asc Description: This is a digitally signed message part.
Re: mdadm without initramfs
Boyd Stephen Smith Jr. b...@iguanasuicide.net writes: In 20100815190053.ga4...@gandalf.home.lxtec.de, Elimar Riesebieter wrote: How do I set up mdadm to create the root array witout an initramfs? You can't. When did this change? I have a box (my NAS) that is running Debian stable (lenny) with a custom 2.6.30 kernel. I never build initramfs images. Currently: $ df / Filesystem 1K-blocks Used Available Use% Mounted on /dev/md0 3842296 1578104 2225156 42% / $ cat /proc/cmdline root=/dev/md0 ro My /dev/sd[bc]1 partitions are of type 0xFD. /dev/md0 has a version 00.90 superblock. If I upgrade this kernel, or upgrade to the new stable when it's released, is this box going to stop booting? -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/2fc2.4c69d2df.bf...@getafix.xdna.net
Re: mdadm without initramfs
also sprach Cameron Hutchison li...@xdna.net [2010.08.17.0207 +0200]: My /dev/sd[bc]1 partitions are of type 0xFD. /dev/md0 has a version 00.90 superblock. If I upgrade this kernel, or upgrade to the new stable when it's released, is this box going to stop booting? Not for now, as long as the md drivers aren't built as modules. But it might stop one day. -- .''`. martin f. krafft madd...@d.o Related projects: : :' : proud Debian developer http://debiansystem.info `. `'` http://people.debian.org/~madduckhttp://vcs-pkg.org `- Debian - when you have better things to do than fixing systems i don't think so, said rene descartes. just then, he vanished. digital_signature_gpg.asc Description: Digital signature (see http://martin-krafft.net/gpg/)
mdadm without initramfs
Hi all, I've set up a new machine with sw raid1. / and /boot are on seperated arrays. raid1 support is compiled direct into 2.6.35.1. The machine is only bootable with an initramfs created by the hints in /usr/share/doc/mdadm/README.upgrading-2.5.3.gz. How do I set up mdadm to create the root array witout an initramfs? Booting the sw raid1 setup without an initramfs ended up with an invalid raid superblock magic on sd[ab]* The setup was booting ext4 before creating the arrays without an intramfs. Thanks for any hints. Elimar -- You cannot propel yourself forward by patting yourself on the back. -- To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100815190053.ga4...@gandalf.home.lxtec.de
Re: mdadm without initramfs
In 20100815190053.ga4...@gandalf.home.lxtec.de, Elimar Riesebieter wrote: How do I set up mdadm to create the root array witout an initramfs? You can't. mdadm is a user-space binary that can't be compiled into the kernel images. Therefore, to run mdadm you need an initramfs, or the file system containing '/' must already be mounted by the kernel. Last I checked, it is possible to have the kernel itself start the root array, via a (series of) kernel command-line arguments. However, this doesn't use mdadm or any of its configuration files. Similarly, mdadm doesn't read the kernel command-line, so it is possible the configurations to diverge. It may be possible to have a GRUB2 hook to generate the kernel command-line arguments from the mdadm configuration, but I've not seen such a hook. In the absence of such a hook (or when you are not using GRUB2 as your bootloader), you should simply use an initramfs. -- Boyd Stephen Smith Jr. ,= ,-_-. =. b...@iguanasuicide.net ((_/)o o(\_)) ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-' http://iguanasuicide.net/\_/ signature.asc Description: This is a digitally signed message part.
Re: mdadm without initramfs
also sprach Boyd Stephen Smith Jr. b...@iguanasuicide.net [2010.08.16.0034 +0200]: Last I checked, it is possible to have the kernel itself start the root array, via a (series of) kernel command-line arguments. However, this doesn't use mdadm or any of its configuration files. That only works for the deprecated v0.9 superblocks and requires the partitions to be marked with type 0xFD. The whole method is deprecated. Use an initramfs. -- .''`. martin f. krafft madd...@d.o Related projects: : :' : proud Debian developer http://debiansystem.info `. `'` http://people.debian.org/~madduckhttp://vcs-pkg.org `- Debian - when you have better things to do than fixing systems welcome to american airlines, sir. here's your avocado - remember to keep it turned on and with you at all times. please turn your luggage over to the armadillos for rootling. -- http://azure.humbug.org.au/~aj/armadillos.txt digital_signature_gpg.asc Description: Digital signature (see http://martin-krafft.net/gpg/)
