[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: LVM root?



On Thu, Oct 12, 2006 at 04:35:02PM +0200, Jean-Luc Coulon (f5ibh) wrote:
> *Real* hardware raid doesnt need an OS layer / driver to work.
> This kind of raid relies on the BIOS *and* on a Windows driver.
> It is more a raid feature enabled in the BIOS and managed by the  
> Windows driver.
> Linux can or not support this BIOS feature depending of the chipset.
> 
> Most of the time, you have to disable the raid in the BIOS and use pure  
> software raid.
> 
> In case if a raid1, if a disk fails you get a message from the system.
> If you have spare disks (configured and installed as so in the raid),  
> the raid is rebuilt on the spare disk. You will notice disk activity  
> related to this mirroring. Then you can wait until you can shutdown  
> your system and remove/replace the defective disk. Then restart the  
> system and use mdadm to reinstall the disk in the array.
> If you have no spare, the raid is degraded and you are running on the  
> safe disk without any redundancy.
> You have, the same way as before, to shutdown your system and remove /  
> replace the defective disk.
>
> In case of a raid0, you have no redundancy and the filesystem relying  
> on this raid will die.
> 
> Remarks :
> - SATA is told to be hotplug but most of the motherboards dont support  
> hotplugging of disks on their SATA controller. This is why you have to  
> shutdown your system.

Actually most SATA controllers DO support hotplug.  The linux kernel
doesn't support SATA hot plug yet (although supposedly that is being
worked on).

> - There are disk failures and there are controller failures. If both  
> your disk are on the same controller, your system will crash.

If your controller fails, most likely the system will crash when the
driver gets very unexpected results.  This is a rather unusual case, and
very expensive to protect against.  Disk failures are much more common
and fortunately much easier to protect against.

> - The swap has to be on the raid also (or on a logical volume of LVM  
> which is built over the raid) otherwise, you will probably crash your  
> system at the failure.

That is for sure.

> You can download and install mdadm : the doc files in  
> /usr/share/doc/mdadm contain valuable informations.
> 
> It is often easier to repair have access to ext3 (which is  
> ext2+journal) from a system you have booted from a live CD, just in  
> case of a weird problem on the filesystem.

Also grub only supports some filesystems for /boot, so it makes sense to
keep / as something supported and simple.  After all if your / is small
and doesn't change much why would you need any fancy filesystem for it?

--
Len Sorensen



Reply to: